[go: up one dir, main page]

WO1999061654A1 - Procedes et systeme de prediction des fonctions biologiques de proteines - Google Patents

Procedes et systeme de prediction des fonctions biologiques de proteines Download PDF

Info

Publication number
WO1999061654A1
WO1999061654A1 PCT/US1999/011911 US9911911W WO9961654A1 WO 1999061654 A1 WO1999061654 A1 WO 1999061654A1 US 9911911 W US9911911 W US 9911911W WO 9961654 A1 WO9961654 A1 WO 9961654A1
Authority
WO
WIPO (PCT)
Prior art keywords
protein
functional site
functional
descriptor
sequence
Prior art date
Application number
PCT/US1999/011911
Other languages
English (en)
Inventor
Jeffrey Skolnick
Jacquelyn S. Fetrow
Adam Godzik
Lukasz Jaroszewski
Leszek Rychlewski
Baohong Zhang
Original Assignee
The Scripps Research Institute
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by The Scripps Research Institute filed Critical The Scripps Research Institute
Priority to AU42185/99A priority Critical patent/AU4218599A/en
Publication of WO1999061654A1 publication Critical patent/WO1999061654A1/fr

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
    • G01N33/50—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing
    • G01N33/68—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing involving proteins, peptides or amino acids
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/30—Detection of binding sites or motifs
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B15/00—ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B15/00—ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
    • G16B15/20—Protein or domain folding
    • C—CHEMISTRY; METALLURGY
    • C07—ORGANIC CHEMISTRY
    • C07K—PEPTIDES
    • C07K2299/00—Coordinates from 3D structures of peptides, e.g. proteins or enzymes
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations

Definitions

  • the present: invention concerns methods and systems for predicting tne function of proteins.
  • the invention relates to materials, software, automated systems, and methods for implementing the same m order to predict the funct ⁇ on(s) of a protein.
  • Protein function prediction includes the use of functional site descriptors for a particular protein function.
  • viruses contain on the order of 10-50 genes and require components supplied by a cell of another organism m order to reproduce, the genomes of independent, living organisms (i.e., those having a genome that encodes for all the information required for the organism to survive and reproduce) that are the least genetically complex have more than 400 genes (for example, Mycoplasma gem taliurn) .
  • More complex, multicellular organisms e.g., mice or humans
  • Most orgamsmal genomes are comprised of double-stranded DNA.
  • Each strand of the genomic DNA is comprised of a long polymer of the four deoxyribonucleotide bases A (adenine) , T (thymme) , G (guanine) , and C (cytosme) .
  • Douole-stranded DNA is formed by the anti-parallel , non-covalent association between two DNA strands. This association is mediated by hydrogen bonding between nucleotide bases, with specific, complementary pairing of A with T and G with C.
  • Each gene m the genomic DNA is expressed by transcription, wherein a single-stranded RNA copy of the gene is transcribed from the double-stranded DNA. The transcribed strand of RNA is complementary to the coding strand of the DNA.
  • RNA is composed of ribonucleotide (rather than deoxyribonucleotide) bases, three of which are similar to those found m DNA: A, G, and C.
  • the fourth RNA ribonucleotide base, uracil (U) substitutes for T found in DNA and is complementary to the A base.
  • RNA molecules for example, ribosomal RNAs, small nuclear RNAs, transfer RNAs, and ribozymes (i.e., RNA molecules having endoribonuclease catalytic activity) .
  • RNAs serve as messengers (mRNAs) , and these are translated into polypeptides.
  • mRNAs messengers
  • the particular sequence of the ribonucleotides incorporated into an RNA as it is synthesized is dictated by the gene found in the genomic DNA from which it was transcribed. In the translation of an mRNA, the particular nucleotide sequence determines the particular amino acid sequence of the polypeptide translated therefrom.
  • each nucleotide triplet, or "codon” (of which there are 4 3 , or 64, possibilities) codes for one amino acid, except that three codons code for no amino acids (each being a "stop" translation codon) .
  • sequence of codons (dictated by the nucleotide sequence of the corresponding gene) specifies the amino acid sequence of a particular protein, and it is the amino acid sequence that ultimately determines the three- dimensional structure of the protein.
  • three- dimensional structure dictates the particular biological function (s) of any biomolecule, including proteins.
  • the haploid human genome comprises about 3 x 10 9 (three trillion) nucleotides spread across 23 chromosomes.
  • the biological function (s) of the gene products encoded by many of the genes sequenced so far remain unknown. Similar situations exist with respect to the genomes of many other organisms.
  • each sequence is located in the genome, and what biological function (s), if any, the sequence encodes, i.e., what is the purpose of the sequence or, if transcribed (or transcribed and translated) , the resulting product, a biological system?
  • the sequence a regulatory region or, if it is transcribed (or transcribed and translated) , does the gene product bind to another molecule, regulate a cellular process, or catalyze a chemical reaction?
  • a number of methods have been developed an attempt to glean functional information about a deduced ammo acid sequence.
  • the most common computational methods include sequence alignment and analysis of local sequence motifs, although these methods are limited by the extent of sequence similarity between sequences of unknown and known function. Additionally, these methods increasingly fail as sequence identity decreases.
  • Other recently developed computational methods include whole genome comparison (Himmelreich et al . , 1997), and analysis of gene clustering (Hiirrmelreich et al . , 1997; Tama es et al . , 1997).
  • Others have developed experimental methods to analyze protein function on a gemone- wide basis These methods include, for example, "two hybrid screens" (Fromont-Racine et al .
  • Sequence alignment is the method most commonly used attempts to identify protein function from am o acid sequence. In this method, the extent of ammo acid sequence identity between an experimental sequence and one or more sequences whose function (s) is (are) known is computed. Alignment methods such as BLAST (Altschul et al . , 1990), BLITZ (MPsrch) (Sturrock & Collins, 1993) , and FASTA (Pearson & Lipman, 1988) are typically employed for this purpose. Assignment of function is based on the theory that significant sequence identity strongly predicts functional similarity (Fitch, 1970?) .
  • Acids Res . , vol. 22:3590-3596 use local sequence information (i.e., the sequence of several contiguous ammo acid residues), as opposed to entire ammo acid sequences, in order to try to identify sequence patterns that are specific for a given function.
  • local sequence motifs cannot recognize conserved three-dimensional structure - motifs can only recognize local sequence. Consequently, local sequence motifs may fail to be accurate predictors of protein function because function derives from three-dimensional structure. In other words, local sequence motif analysis is limited where function is dependent upon non-local residues, i.e., ammo acids disposed m different regions of a protein's primary structure.
  • Root mean square distances were then determined for all Ser and Asp ° side chain atoms m a set of serme proteases whose structures were also then known at atomic resolution. This analysis revealed that the positioning of a single oxygen atom each of the Asp and Ser side chains was highly conserved.
  • a 3D template was developed for serme protease 5 activity using the identity of three ammo acids, namely Ser, His, and Asp, and the 3D coordinate positions (to an RMSD cutoff of 2 A) for the functional oxygen atoms m the Ser and Asp side chains and the non-hydrogen atoms of the His side chain. The 3D template was then applied to a test set of high 0 resolution protein structures drawn from the PDB database.
  • the inventions described and claimed herein solve these needs by providing novel methods and systems for predicting protein function from sequence.
  • Various methods described and claimed herein use sequence and structure information and apply this information to protein structures, particularly inexact models of protein structure, that can be computationally derived from protein or nucleic acid sequences. Using their methods, the inventors have discovered that it is not necessary to accurately predict the overall three-dimensional structure of a particular protein of interest m order to predict its function. Instead, prediction of biological function using the methods described and claimed herein requires only an approximation of the three-dimensional orientation of two or more ammo acid residues m a region responsible for the particular function of the protein under investigation.
  • this invention overcomes the problems and limitations of the methods previously utilized m an attempt to identify protein function from either sequence or structure. As those m the art will appreciate, such methods can routinely be adapted with respect to various protein functional sites m order to predict protein function. A more detailed description of the invention is provided below. 3. Definitions
  • an “ammo acid” is a molecule (see Figure 1) having the structure wherein a central carbon atom (the alpha ( ⁇ ) -carbon atom) is linked to a hydrogen atom, a carboxylic acid group (the carbon atom of which is referred to herein as a “carboxyl carbon atom”) , an ammo group (the nitrogen atom of which is referred to herein as an "ammo nitrogen atom”), and a side chain group, R.
  • an ammo acid loses one or more atoms of its ammo and carboxylic groups the dehydration reaction that links one ammo acid to another.
  • an ammo acid when incorporated into a protein, an ammo acid is referred to as an "ammo acid residue."
  • an ammo acid residue's R group differentiates the 20 ammo acids from which proteins are synthesized, although one or more ammo acid residues m a protein may be derivatized or modified following incorporation into protein m biological systems (e.g., by glycosylation and/or by the formation of cystme through the oxidation of the thiol side chains of two non-adjacent cysteme ammo acid residues, resulting m a disulfide covalent bond that frequently plays an important role m stabilizing the folded conformation of a protein, etc.) .
  • non-naturally occurring ammo acids can also be incorporated into proteins, particularly those produced by synthetic methods, including solid state and other automated synthesis methods.
  • ammo acids include, without limitation, ⁇ -ammo isobutyric acid, 4-ammo butyric acid, L-ammo butyric acid, 6-ammo hexanoic acid, 2 -ammo isobutyric acid, 3 -ammo propionic acid, ornithine, norlensme, norvalme, hydroxprolme, sarcos e, citrall e, cysteic acid, t-butylglyme, t-butylalanme, phenylylycme, cyclohexylalamne, ⁇ -alanme, fluoro-ammo acids, designer ammo acids (e.g., ⁇ -methyl ammo acids, ⁇ -methyl ammo acids, N ⁇ -methyl ammo
  • Non-polar (hydrophobic) ammo acids include alamne , leucme, isoleucine, valme, prolme, phenylalanine, tryptophan, and methionmes .
  • Neutral ammo acids include glycme, serine, threonine, cysteme, tyrosine, esparagme, and glutamme.
  • Positively charged (basic ammo acids include argmme, lysine and histidine.
  • Negatively charged (acidic) ammo acids include aspartic acid and glutamic acid.
  • a " ⁇ -carbon atom” refers to the carbon atom (if present) m the R group of the side chain of an ammo acid (or ammo acid residue) that is covalently bonded to the ⁇ -carbon atom of that ammo acid (or residue) . See Figure 1.
  • glycme is the only naturally occurring am o acid found m mammalian proteins that does not contain a ⁇ -carbon atom.
  • a “biomolecule” refers to any molecule (including synthetic molecules) produced by a cell, found withm a cell or organism, or which can be introduced into a cell or organism, or any derivative of such a molecule, and any other molecule capable of performing or having a biological function.
  • Representative examples of biomolecules include nucleic acids and proteins.
  • a “synthetic" biomolecule is one that has been prepared, m whole or part, through the use of one or more synthetic chemical reactions.
  • Protein refers to any polymer of two or more individual ammo acids (whether or not naturally occurring) linked via a peptide bond, and occurs when the carboxyl carbon atom of the carboxylic acid group bonded to the ⁇ -carbon of one ammo acid (or ammo acid residue) becomes covalently bound to the ammo nitrogen atom of ammo group bonded to the ⁇ -carbon of an adjacent ammo acid. See Figure 1.
  • polypeptide backbone of the protein.
  • the polypeptide backbone shall be understood to refer the ammo nitrogen atoms, ⁇ -carbon atoms, and carboxyl carbon atoms of the protein, although two or more of these atoms (with or without their substituent atoms) may also be represented as a pseudoatom.
  • any representation representing a polypeptide backbone that can be used a functional site descriptor as described herein will be understood to be mcluded withm the meaning of the term "polypeptide backbone.”
  • protein is understood to include the terms “polypeptide” and “peptide” (which, at times, may be used interchangeably herein) withm its meaning.
  • proteins comprising multiple polypeptide subunits e.g., DNA polymerase III, RNA polymerase II
  • other components for example, an RNA molecule, as occurs m telomerase
  • fragments of proteins and polypeptides are also withm the scope of the invention and may be referred to herein as “proteins .
  • the particular ammo acid sequence of a given protein i.e., the polypeptide' s "primary structure," when written from the amino- erminus to carboxy- termmus
  • the particular ammo acid sequence of a given protein is determined by the nucleotide sequence of the coding portion of a mRNA, which is m turn specified by genetic information, typically genomic DNA (which, for purposes of this invention, is understood to include organelle DNA, for example, mitochondrial DNA and chloroplast DNA) .
  • any type of nucleic acid which constitutes the genome of a particular organism e.g., double-stranded DNA m the case of most animals and plants, single or douole-stranded RNA m the case of some viruses, etc.
  • Messenger RNA is translated on a ⁇ bosome, which catalyzes the polymerization of a free ammo acid, the particular identity of which is specified by the particular codon (with respect to mRNA, three adjacent A, G, C, or U ribonucleotides m the mRNA' s coding region) of the mRNA then being translated, to a nascent polypeptide.
  • Recombinant DNA techniques have enabled the large-scale synthesis of polypeptides (e.g., human insulin, human growth hormone, erythropoietm, granulocyte colony stimulating factor, etc.) having the same primary sequence as when produced naturally m living organisms.
  • polypeptides e.g., human insulin, human growth hormone, erythropoietm, granulocyte colony stimulating factor, etc.
  • such technology has allowed the synthesis of analogs of these and other proteins, which analogs may contain one or more ammo acid deletions, insertions, and/or substitutions as compared to the native proteins.
  • Recombinant DNA technology also enables the synthesis of entirely novel proteins.
  • the primary structure of a protein (which also includes disulfide (cystine) bond locations) can be determined by the user.
  • polypeptides having a primary structure that duplicates that of a biologically produced protein can Oe achieved, as can analogs of such proteins.
  • completely novel polypeptides can also be synthesized, as can protein incorporating non-naturally occurring ammo acids.
  • the peptide bonds between adjacent ammo acid residues are resonance hybrids of two different electron isomeric structures, wherein a bond between a carbonyl carbon (the carbon atom of the carboxylic acid group of one ammo acid after its incorporation into a protein) and a nitrogen atom of the ammo group of the ⁇ -carbon of the next ammo acid places the carbonyl carbon approximately 1.33 A away from the nitrogen atom of the next ammo acid, a distance about midway between the distances that would be expected for a double bond (about 1.25 A) and a single bond (about 1.45 A) .
  • This partial double bond cnaracter prevents free rotation of the carbonyl carbon and ammo nitrogen about the bond thereoetween under physiological conditions.
  • the atoms bonded to the carbonyl carbon and ammo nitrogen reside in the same plane, and provide discrete regions of structural rigidity, and hence conformational predictability, m proteins.
  • each ammo acid residue contributes two additional single covalent bonds to the polypeptide chain.
  • a helix pitch refers to the distance between repeating turns on a line drawn parallel to the helix axis. Bond angles associated with other secondary structures are known m the art, or can be determined experimentally using standard techniques .
  • proteins also have secondary, tertiary, and, in multisubunit proteins, quaternary structure.
  • Secondary structure refers to local conformation of the polypeptide chain, with reference to the covalently linked atoms of the peptide bonds and ⁇ -carbon linkages that string the ammo acids of the protein together. Side chain groups are not typically included m such descriptions.
  • Representative examples of secondary structures include ⁇ helices, parallel and anti-parallel ⁇ structures, and structural motifs such as helix-turn-helix, ⁇ - ⁇ - ⁇ , the leucme zipper, the zmc finger, the ⁇ -barrel, and the immunoglobulin fold.
  • Tertiary structure concerns the total three-dimensional structure of a protein, including the spatial relationships of ammo acid side chains and the geometric relationship of different regions of the protein. Quaternary structure relates to the structure and non-covalent association of different polypeptide subunits m a multisubunit protein.
  • a “functional site” refers to any site in a protein that has a function. Representative examples include active sites (i.e., those sites m catalytic proteins where catalysis occurs), protem-protem interaction sites, sites for chemical modification (e.g., glycosylation and phosphorylation sites), and ligand binding sites.
  • Ligand binding sites include, but are not limited to, metal binding sites, co- factor binding sites, antigen binding sites, substrate channels and tunnels, and substrate binding sites. In an enzyme, a ligand binding site that is a substrate binding site may also be an active
  • a "pseudoatom” refers to a position m three dimensional space (represented typically by an x, y, and z coordinate set) that represents the average (or weighted average) position of two or more atoms in a protein or ammo acid.
  • Representative examples of a pseudoatom include an ammo acid side chain center of mass and the center of mass (or, alternatively, the average position) of an ⁇ -carbon atom and the carboxyl atom bonded thereto.
  • a “reduced model” refers to a three-dimensional structural model of a protein wherein fewer than all heavy atoms (e.g., carbon, oxygen, nitrogen, and sulfur atoms) of the protein are represented.
  • a reduced model might consist of just the ⁇ -carbon atoms of the protein, with each ammo acid connected to the subsequent ammo acid by a virtual bond.
  • Other examples of reduced protein models include those m which only the ⁇ -carbon atoms and side chain centers of mass of each ammo acid are represented, or where only the polypeptide backbone is represented.
  • a “geometric constraint” refers to a spatial representation of an atom or group of atoms (e.g., an ammo acid, the R-group of an ammo acid, the center of mass of an R-group of an ammo acid, a pseudoatom, etc.) . Accordingly, such a constraint can be represented by coordinates in three dimensions, for example, as having a certain position, or range of positions, along x, y, and z coordinates (i.e., a "coordinate set”) .
  • a geometric constraint can be represented as a distance, or range of distances, between a particular atom (or group of atoms, etc.) and one or more other atoms (or groups of atoms, etc.) .
  • Geometric constraints can also be represented by various types of angles, including the angle of bonds (particularly covalent bonds, e.g., ⁇ bonds and ⁇ bonds) between atoms m an ammo acid residue, between atoms m different ammo acid residues, and between atoms m an ammo acid residue of a protein and another molecule, e.g. , a ligand, with ranges for each angle being preferred.
  • a “conformational constraint” refers to the presence of a particular protein conformation, for example, an ⁇ -helix, parallel and antiparallel ⁇ strands, leucme zipper, zmc finger, etc.
  • conformational constraints can include ammo acid sequence information without additional structural information.
  • “-C-X-X-C-” is a conformational constraint indicating that two cysteme residues must be separated by two other ammo acid residues, the identities of each of which are irrelevant the context of this particular constraint.
  • identity constraint refers to a constraint of a functional site descriptor that indicates the identity of an ammo acid residue at a particular location m a protein. (determined by counting the number of ammo acid residues m the protein from its ammo terminus up to and including the residue m question) . As those m the art will appreciate, comparison between related proteins may reveal that the identity of a particular ammo acid residue at a given ammo acid position m a protein is not entirely conserved, i.e., different ammo acid residues may be present at a particular ammo acid position m related proteins.
  • two or more alternative ammo acid residue identities can be provided for a particular identity constraint of a functional site descriptor.
  • the invention also envisions different functional site descriptors for the particular biological function that differ by employing different ammo acid residue identities (or sets of identities) for the corresponding position. For example, where it is determined by sequence alignment that related proteins have one of two ammo acid residues at a particular position m the functional site, a single functional site descriptor therefor may specify the two alternatives. Alternatively, two different functional site descriptors may be generated that differ only with respect to the identity constraint at that position.
  • a functional site descriptor refers to the inclusion of a user-defined variance therein.
  • the degree of relaxation will depend on the particular constraint and its application. As those m the art will appreciate, functional site descriptors for the same biological function can be developed wherein different degrees of relaxation for one or more constraints are what differentiate one such descriptor from another.
  • Protein structures useful the practice of the invention can oe of different quality.
  • the highest quality determination methods are experimental structure prediction methods based on x-ray crystallography and NMR spectroscopy.
  • high resolution structures are those wherein atomic positions are determined at a resolution of about 2 A or less, and enable the determination of the three-dimensional positioning of each atom (or each non- hydrogen atom) of a protein.
  • “Medium resolution” structures are those wherein atomic positioning is determined at about the 2-4 A level, while “low resolution” structures are those wherein the atomic positioning is determined m about the 4-8 A range.
  • protein structures that have been determined by x-ray crystallography or NMR may be referred to as "experimental structures,” as compared to those determined by computational methods, i . e . , derived from the application of one or more computer algorithms to a primary ammo acid sequence to predict protein structure.
  • protein structures can also be determined entirely by computational methods, including, but not limited to, homology modeling, threading, and ab mi tio methods.
  • models produced by such computational methods are "reduced” models, i . e .
  • the predicted structures do not include all non-hydrogen atoms in the protein. Indeed, many reduced models only predict structures that snow the polypeptide backbone of the protein, and such models are preferred in the practice of the invention.
  • a protein structure based on a reduced model has been generated, all or a portion of it may be further refined to include additional predicted detail, up to including all atom positions.
  • Computational methods usually produce lower quality structures than experimental methods, and the models produced by computational methods are often called “inexact models.” While not necessary m order to practice the instant methods , the precision of these predicted models can be determined using a benchmark set of proteins whose structures are already known The predicted model for each biomolecule may then be compared to a corresponding experimentally determined structure. The difference between the predicted model and the experimentally determined structure is quantified v a a measure called "root mean square deviation" (RMSD) . A model having an RMSD of about 2.0 A or less as compared to a corresponding experimentally determined structure is considered “high quality" . Frequently, predicted models have an RMSD of about 2.0 A to about 6.0 A when compared to one or more experimentally determined structures, and are called
  • RMSDs can also be determined for one or more atomic positions when two or experimental structures have been generated for the same protein.
  • the object of this invention is to enable one or more functions of a protein to be predicted from structural information, for example, from computationally derived models of protein structure (including inexact models) produced from deduced primary ammo acid sequences, for example, as may be derived from nucleotide sequence of a novel gene obtained m the course of genome sequencing projects.
  • the present invention comprises a number of objects, aspects, and embodiments.
  • One aspect of the invention concerns functional site descriptors (FSDs) that define spatial configurations for protein functional sites that correspond with particular biological functions. It is known that function derives from structure.
  • a functional site descriptor according to the invention provides three-dimensional representation of protein functional site.
  • the functional site represented by an FSD is a ligand binding domain (e.g., a domain that binds a ligand, for example, a substrate, a co- factor, or an antigen), while m other embodiments, the functional site is a protem-protem interaction site or domain.
  • the functional site is an enzymatic active site. Particularly preferred embodiments concern functional sites other than a divalent metal ion binding site.
  • a functional site descriptor typically comprises a set of geometric constraints for one or more atoms each of two or more ammo acid residues comprising a functional site of a protein.
  • at least one of said two or more ammo acid residues is also identified as a particular ammo acid residue or set of ammo acid residues.
  • the said one or more atoms is selected from the group consisting of amide nitrogens, ⁇ -carbons, carbonyl carbons, and carbonyl oxygens withm a polypeptide backbone, ⁇ -carbons of ammo acid residues, and pseudoatoms .
  • at least one of said one or more atoms is an amide nitrogen, an ⁇ -carbon, a ⁇ -carbon, or a carbonyl oxygen withm a polypeptide backbone.
  • a functional site descriptor represents 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, or 15 ammo acid residues (or sets of residues) that comprise the corresponding the functional site.
  • an FSD may include one or more identity constraints with respect to any ammo acid, such constraints preferably make reference to naturally occurring ammo acids, particularly naturally occurring L ammo acids, including those selected from the group consisting of Ala, Arg, Asn, Asp, Cys, Gin, Glu, Gly, His, lie, Leu, Lys, Met, Phe, Pro, Ser, Thr, Trp, Tyr, and Val .
  • the geometric constraints of an FSD preferably are selected from the group consisting of an atomic position specified by a set of three dimensional coordinates, an interatomic distance (or range of interatomic distances) , and an interatomic oond angle (or range of interatomic bond angles) .
  • a geometric constraint refers to atomic position
  • reference is typically made to a set of three dimensional coordinates.
  • Such constraints preferably relate to RMSDs, particularly those that allow the atomic position to vary with a preselected RMSD, for example, by an amount of less than about 3 A, less than about 2.5 A, less than about 2.0 A, less than about 1.5 A, and less than about 1.0 A.
  • Other geometric constraints concern interatomic distances, preferably interatomic distance ranges, or interatomic bond angles range preferably interatomic bond angle ranges.
  • an FSD can also include one or more conformational constraints that refer to tne presence of a particular secondary structure, for example, a helix, or location, for example, near the ammo or carboxy terminus of a protein.
  • an FSD refers to at least one atom from each of 4 , 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, or 15 ammo acid residues that comprise the corresponding functional site.
  • all of the atoms for which geometric constraints are provided comprise a part of the polypeptide backbone and are selected from the group consisting of ⁇ -carbons, amide nitrogens, carbonyl carbons, and carbonyl oxygens.
  • one or more of such atoms can be a pseudoatom.
  • Representative examples of pseudoatoms are centers of mass, such as may be derived from at least two atoms, such as two or more atoms from one ammo acid residue or two or more atoms from at least two ammo acid residues of the protein.
  • FSDs are those comprising multiple geometric constraints.
  • Representative examples of such FSDs are a three atom functional site descriptor, a four atom functional site descriptor, a five atom functional site descriptor, a six atom functional site descriptor, a seven atom functional site descriptor, an eight atom functional site descriptor, a nine atom functional site descriptor, a ten atom functional site descriptor, an eleven atom functional site descriptor, a twelve atom functional site descriptor, a thirteen atom functional site descriptor, a fourteen atom functional site descriptor, and a fifteen atom functional site descriptor.
  • FSDs according to the invention are implemented m electronic form.
  • libraries of FSDs m electronic or other form.
  • a library comprises at least two functional site descriptors for at least one of the biological functions represented by the library.
  • Another aspect of the invention concerns methods of identifying a protein as having a particular biological function. Sucn methods may also be referred to as function screening methods. Typically, such methods comprise applying a functional site descriptor according to the invention to a structure of a protein and determining whether the protein has the biological function. This determination is made if application of the functional site descriptor reveals that a portion of the structure of the protein matches, or satisfies, the constraints of the functional site descriptor.
  • the structure (s) to which one or more FSDs is (are) applied is (are) of high resolution.
  • High resolution structures can be obtained by a variety of methods, including x-ray crystallography and nuclear magnetic resonance.
  • FSDs can be applied to a protein structures derived from any organism, be they prokaryotic or eukaryotic.
  • Prokaryotic organisms the proteins of which may be screened according to the instant methods include bacteria.
  • Eukaryotic organisms include plants and animals, particularly those of medical or agricultural import.
  • a representative class is mammals, including bovine, canine, equine, feline, ovine, porcine, and primate animals, as well as humans. The methods may also be applied to study viral protein function.
  • the methods of the invention are practiced using plurality of functional site descriptors and/or a plurality of proteins structures, of the same or different proteins, preferably to a plurality of structures for a plurality of proteins.
  • Another aspect of the invention concerns methods of making FSDs for functional sites of proteins (other than divalent metal ion binding sites) , which FSDs can then be applied to inexact, three dimensional structural proteins models .
  • Yet another aspect concerns computer program products comprising a computer useable medium having computer program logic recorded thereon for creating a functional site descriptor for use in predicting a biological function of a protein.
  • Such computer program logic preferably comprises computer program code logic configured to perform a series of operations, including determining a set of geometric constraints for a functional site associated with a biological function of a protein; modifying one or more geometric constraints of said set of geometric constraints to produce a modified set of geometric constraints; comparing said modified set of geometric constraints to a data set of functional sites correlated with said biological function to determine whether said modified set of geometric constraint compares favorably with said data set of functional sites correlated with said biological function and, if so; comparing said modified set of geometric constraint (s) to a data set of functional sites not correlated with said biological function to determine whether said modified set of geometric constraints compares favorably with said data set of functional sites not correlated with said biological function and, if so; repeating said modifying and comparing operations to modify one or more of said geometric constraints of said set of geometric constraints to an extent that said modified set of geometric constraints compares favorably with said data set of functional sites correlated with said biological function without encompassing a predetermined amount of data sets not correlated with said biological function.
  • the operation of determining a set of geometric constraints of a functional site correlated with a biological function of a protein comprises receiving said set of geometric constraints from at least one of the group of a data set of predetermined geometric constraints or from user input.
  • a predetermined variance can be associated with one or more of the geometric constraints to adjust the same.
  • the operation of modifying one or more geometric constraints of said set of geometric constraints to produce a modified set of geometric constraints comprises computing an average value for a geometric constraint within the set of geometric constraints by determining values for said geometric constraint from two different proteins having functional sites that correlate with said biological function, and calculating said average value; computing a standard deviation with respect to such geometric constraint; and applying a multiplier to said computed standard deviation to generate said modified geometry.
  • FIG. 1 illustrates the structure and geometry of a polypeptide backbone. All atoms between the two ⁇ -carbon atoms of two adjacent ammo acid residues are shown. The peptide bond between the carbonyl carbin of one amino acid residue and the amide nitrogen of the adjacent amino acid is shaded. The dimensions and bond angles shown are averages observed crystallographically in amino acids and shall peptides. (Ramachandran, et al . (1974) Biochim. Biophys .
  • FIG. 2 is an operational flow diagram generally illustrating a process for preparing a functional site descriptor for a given molecular function, ano using the functional site descriptor to determine whether a given molecular model should exhibit the functional properties associated with the functional site descriptor according to one embodiment of the invention.
  • FIG. 3 is a functional block diagram generally illustrating functionality that can be associated with the various processes and techniques for generating functional site descriptors and using functional site descriptors to predict molecular function according to one embodiment of the invention.
  • FIG. 4 is an operational flow diagram illustrating a process for creating a functional site descriptor for a given function according to one embodiment of the invention.
  • FIG. 5 is an operational flow diagram illustrating a method for creating a functional site descriptor according to one embodiment of the invention.
  • FIG. 6, which comprises FIGS. 6A, 6B, and 6C, illustrates residues of a hypothetical active site descriptor (designated by numbers 3, 84, 142), adjacent residues (designated by numbers 2 and 4 for the residue designated as 3 , 83 and 85 for the residue designated as 84, and 141 and 143 for the residue designated as 142), and geometries (here, distance ranges, indicated by dotted and hatched lines) among the residues according to one embodiment of the invention.
  • Residue 84 is the primary residue and residue 3 is the secondary residue m Figures 6A and 6B.
  • the primary and secondary residues are more generally designated as "A" and "B” , respectively, as shown Figure 6C.
  • FIG. 7 is an operational flow diagram illustrating one example process for adjusting distance geometries according to one embodiment of the invention.
  • FIG. 8 is a diagram illustrating an example data set for a functional site descriptor according to one embodiment of the invention.
  • A is the primary residue
  • B is the secondary residue
  • C is the tertiary residue, as also illustrated in Figure 6C.
  • FIG. * s an operational flow diagram illustrating selection of a multiplier to be used in determining how much relaxation can be used with regard to one or more of the geometries (e.g., distance range) used in a given functional site descriptor, for example, an active site descriptor.
  • FIG. 10 is an operational flow diagram illustrating a process for utilizing one or more functional site descriptors to determine whether a nucleotide or am o acid sequence encodes such function (s).
  • FIG. 11 is a block diagram illustrating an example processor-based system according to one embodiment of the invention.
  • FIG. 12 shows an alignment of 12 ammo acid sequences, as well as the constraint sets for an FSD.
  • one or more functional site descriptors can be- developed for a biomolecule functional site having a specific biological function.
  • Libraries of such descriptors (preferably implemented electronic form) can be used to probe for or evaluate the activity or function associated with the functional site descriptor in one or more protein structures.
  • protein structures may be represented as inexact structural models.
  • the instant invention has many applications.
  • the invention can be used to identify or screen for a novel function in one or more proteins, to confirm a previously identified or suspected function of a protein, or to provide further information about a specific functional site m a protein.
  • additional information includes, but is not limited to, information about specific active site residues, substrate specificity, or regulatory processes.
  • Functional site descriptors are defined based on various constraints (or “parameters”), including, but not limited to, atomic or molecular geometry and residue identity. Preferably, such parameters are derived from functional sites of proteins of known, high-resolution structure.
  • the descriptors of the invention thus enable identification of functional sites not only from experimentally determined structures, but, significantly, also from predicted structures, including inexact protein models predicted by ah mi tio folding algorithms, threading algorithms, homology modeling algorithms, or other protein structure prediction algorithms .
  • Functional site descriptors define a spatial configuration for a protein functional site that corresponds to a biological function.
  • Functional site descriptors according to the invention comprise, at a minimum, a spatial representation or configuration of at least tv/o atoms, or groups of atoms.
  • a functional site descriptor may be prepared using the interatomic distance, or preferably, a range of interatomic distances, between the ⁇ - carbon atoms of two ammo acid residues known or suspected to be involved m the catalysis carried out by a particular enzyme.
  • such a configuration can be represented in three dimensions using x, y, and z coordinates to identify the position, or range of positions, that a particular atom may have relative to other functionally important residues.
  • each functionally important ammo acid residue, distance (or range of distances) between atoms or pseudoatoms, coordinate set, or other parameter represents a constraint with respect to the particular functional site descriptor.
  • a functional site descriptor will include one or more identity constraints, for example, the identity of a particular ammo acid residue (cr set of am o acid residues) located or predicted to be located at a particular position m a protein, m addition to a set of two or more geometric constraints.
  • identity constraints for example, the identity of a particular ammo acid residue (cr set of am o acid residues) located or predicted to be located at a particular position m a protein, m addition to a set of two or more geometric constraints.
  • other information can also be included for a particular functional site descriptor, for example, information regarding bond angles (or bond angle ranges) , secondary structure information, ammo acid sequence, etc. Whether, and to what extent, such or other information is included a particular functional site descriptor is with
  • RMSDs for one or more atoms or pseudoatoms of each of two or more ammo acid residues identified as comprising at least a portion of the functional site corresponding to the particular functional site descriptor.
  • Other preferred embodiments utilize interatomic distances (particularly distance ranges) between atoms (or pseudoatoms) of different ammo acid residues as geometric constraints.
  • a functional site descriptor comprises a mixture of geometric constraints, some of which are based on RMSDs, while others address interatomic distance ranges. Indeed, the particular composition of constraints m a given functional site descriptor is left to the discretion of the artisan.
  • a functional site descriptor can be prepared as follows. For example, it is preferred to identify at least two, and preferably three or more (e.g., 3, 4, 5 ,6 ,7 ,8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20), ammo acid residue positions, and the ammo acid residue (s) at those positions, known or suspected to be involved m the enzyme's catalytic activity or maintenance of the active site.
  • three or more e.g., 3, 4, 5 ,6 ,7 ,8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20
  • An RMSD defines the three dimensional location of a particular atom or pseudoatom withm a particular range m one or more of x, y, and z coordinates.
  • RSMDs useful m the practice of the invention are most frequently less than about 10 A, preferably less than about 5 A, even more preferably less than about 3 A, and even more preferably less than about 2.5 A to about 0.5 A. While RSMDs as small as about 0.1 A can be used, they are less preferred.
  • the geometric constraints used to define spatial representations of atoms or pseudoatoms of a functional site descriptor utilize distances between one or more atoms of key ammo acid residues relative to each other.
  • a distance constraint between two ⁇ -carbon atoms will typically include the average distance between the two atoms as well as the standard deviation for such distance.
  • the range of distances for any particular spatial or geometric constraint is maximized, but preferably only to the extent that specificity toward identification of the desired functional sites is maintained.
  • the standard deviation for a particular distance or other geometric constraint can be increased by a multiple.
  • a user-defined, pre-selected multiple selected from the group consisting of 0.5 to 10 m increments of 0.1 may be applied to one or more geometric constraints of a functional site descriptor.
  • distance and other spatial or geometric constraints be represented as ranges (i.e., an RMSD or minimum and maximum allowed distance between two atoms, molecules, residues, side chains, or other positions in coordinate space) .
  • a functional site descriptor for an active site of an enzyme may comprise one, and preferably two, three, or more primary residue identity constraints (e.g., the allowed identities of three ammo acids having a role m the enzyme's catalytic function) .
  • Such a descriptor might also be composed of at least three distance constraints, namely the distance ranges between the three ammo acids comprising the primary residue identity constraints.
  • additional distance and/or residue identity constraints may be employed.
  • functional site descriptors comprising two to about 50 or more geometric constraints can be developed for a particular biological function.
  • the number of geometric constraints m a functional site descriptor is from about 4-25, often from about 5-20.
  • a so-called "9-d ⁇ stance" constraint set for an enzyme active site that correlates with a biological function can be employed m conjunction with three primary residue identity constraints (illustrated, for example, m Figure 5) .
  • the functional site descriptor comprises the distance ranges between the ⁇ -carbon atoms of the three ammo acids making up the primary residue identity constraints (three distance constraints) and six additional distance ranges.
  • the three primary residues are designated A, B, and C for purposes of the geometric constraints.
  • the additional distance ranges could be determined by identifying the positions of the ⁇ - carbon atoms of the ammo acids immediately before and after each of the three ammo acids of the primary residue identity constraint according to primary sequence, as illustrated in Figure 5.
  • the additional residues could be designated as A-l and A+l, B-l and B+l, and C-l and C- ⁇ -1, for residues that immediately precede and follow residues A, B, and C (as determined by ammo acid position in the primary sequence of the protein), respectively.
  • the additional residues could be designated as A-l and A+l, B-l and B+l, and C-l and C- ⁇ -1, for residues that immediately precede and follow residues A, B, and C (as determined by ammo acid position in the primary sequence of the protein), respectively.
  • six or more different distance ranges are determined between these secondary ⁇ -carbons and the primary ⁇ -carbons and/or other of the secondary ⁇ -caroons. If two or more structures are examined for purposes of optammg these distances, it s often useful to calculate the standard deviation these distances for purposes of establishing distance ranges.
  • the ranges are increased by applying a multiplier to the upper and/or lower end of the calculated standard deviation range.
  • This multiplier can be any number, although it preferably is between about 0.5 and 10. Often, a suitable multiplier will be 2.0.
  • the number of constraints included m a particular functional site descriptor will vary depending upon the number desired or required to distinguish the particular descriptor from descriptors ,or structures) for, or which correlate with, other biological functions.
  • a functional site descriptor according to the invention will comprise the fewest constraints possible to achieve the selectivity and specificity desired for the particular functional site descriptor.
  • a functional site descriptor will allow all proteins having the particular biological function that corresponds to the descriptor to be identified m the set of proteins being examined ("selectivity"), without identifying proteins that do not have the function ("specificity") .
  • Functional site descriptors according to the invention can be produced, for example, by the following steps, as outlined m Figure 4. Initially, the functionally and structurally important residues are identified (step 212) .
  • proteins are selected that possess the particular function and for which an experimentally determined three dimensional structure
  • the putative functionally important residues are identified m the Known three-dimensional structure (s) , and the relative geometries ⁇ e . g. , distances, angles) between atoms or group of atoms, for example, the ⁇ -carbons of each of the functionally important ammo acids of a protein, are recorded.
  • other structural information such as secondary structural information, may also be included m the descriptor if there is evidence for the importance of such information.
  • Structural superposition and multiple sequence alignment may help identify other residues that might be important m the particular biological function under consideration, but these procedures are generally used only where experimental evidence suggests a functional significance.
  • the functional s te descriptor building procedure is iterative. After identification of conserved residues (or sets of residues) , another analysis may be performed to determine the relative functional importance of the conserved residues (or sets of residues) withm the structures. As those m the art will appreciate, m a functional site descriptor for a protein, the aim is to use only those residues known to be functionally important or conserved across one or more proteins exhibiting the function or activity of interest.
  • the descriptor should be sensitive enough to enable the identification of corresponding functional sites m biomolecules while at the same time having the specificity to detect only those sites and not others having a similar configuration but lacking the desired biological function m other biomolecules.
  • a set of identity, geometric, and/or conformational constraints (it being understood that other constraint types can also be mcluded, if desired, a particular descriptor) for a specific function have been determined, they are preferably implemented m electronic form as described below.
  • One or more of these descriptors can then be used to probe (preferably via computer) one or more experimentally determined protein structures for sets of residues that satisfy, or match, the specified constraints.
  • Experimentally determined protein structures could be, for example, those determined by x-ray crystallography or NMR spectroscopy, which might be stored m a repository, such as Broo haven PDB (Abola et al . , 1987).
  • the constraints are preferably implemented stepwise, so that structures that are eliminated by each criterion may be evaluated at each step along the way. If a constraint set adopted as a functional site descriptor for a particular biological function misses or fails to identify any protein withm the experimentally determined structures known or suspected to exnibit tne particular biological function under investigation, the structure of the missed protein, particularly the structure of the functional site correlated with the desired biological function, is analyzed.
  • tne constraints comprising the functional site descriptor is then modified or relaxed so that the biomolecule exhibiting the function is correctly identified upon re-testing.
  • "Relaxation" of a functional site descriptor constraint or parameter e.g., the distance between the ⁇ -carbons of two functionally important ammo acids, refers to the range of numbers embodied by the particular parameter. Ideally, such relaxation should not degrade the performance of the functional site descriptor m identifying proteins having the desired biological function on high resolution structures.
  • the constraint set for the descriptor is modified to include such a distance.
  • one or more or all members of the constraint set are adjusted to allow, m the case of constraints (e.g., distance constraints, bond angle constraints, etc.), the greatest possible latitude between relevant atoms (or groups of atoms, etc.) without leading to the identification of proteins lacking the desired function but exhibiting a related structure.
  • a particular functional site descriptor is determined to have the requisite selectivity and specificity for its target biological function, it may be still be desirable to further relax one or more of the constraints of the descriptor, so long s the desired level of selectivity and specificity are maintained.
  • m application to further experimentally determined or modeled structures t may be necessary or desirable to further relax one or more constraints m the descriptor to accommodate ambiguities mnerent m medium, low-resolution, or inexact models.
  • the extent of parameter relaxation may be ascertained by testing against various structures, for example, against exact structures (i.e., high-resolution experimentally determined models) and against inexact, predicted models of protein structures.
  • a functional site descriptor it is desirable to relax one or more parameters of a functional site descriptor to the maximum extent possible, other words, to identify all biomolecules of known structure having the desired function without identifying a oiomolecule known not to have ⁇ ,cr that does not have) the desired function.
  • existing functional site descriptors can be probed against sucn structures, and refined, or relaxed (or perhaps both, on a parameter-by-parameter basis) , if needed or desired. If the functional site descriptor selects one or more proteins not known to display the function, then the structure of any such "false positive" example is compared to the known functional sites.
  • the functional descriptor has identified a functional site that was not previously recognized m the protein.
  • the functional descriptor has incorrectly identified a functional site.
  • one or more constraints of the functional site descriptor are modified (for example, m the case of distance constraints, one or more distance constraints is adjusted by altering the interatomic distance ranges between relevant atoms or groups or groups of atoms to eliminate the false positives.
  • sequence, structural, and experimental analysis are utilized to distinguish between these two cases.
  • a putative functional site descriptor is generated.
  • the descriptor is validated for selectivity and specificity using a test set of proteins, some of which possess the biological function corresponding to the descriptor, and some of whicn do not.
  • This descriptor may be applied to experimentally determined or modeled structures.
  • each protein assumes a "native conformation," a unique secondary and tertiary (and quaternary conformation m the case of multisupunit proteins) conformation dictated by the protein's primary structure.
  • the folding of a protein typically is spontaneous and under the control of non-covalent forces, and results m the lowest free energy state kmetically available under the particular pH, temperature, and ionic strength conditions. Disulfide bonds are typically formed after folding occurs, and serve to stabilize the native conformation.
  • proteins having unrelated biological function or sequence can have similar patterns of secondary structure m the tertiary structure of different domains.
  • Non-covalent interactions are weak bonding forces having bond strengths that range from aoout 4 to about 29 kcal/mol, which exceed the average kinetic energy of molecules at 37°C (about 0.6 kcal/mol) .
  • covalent bonds have bond strengths of least about 50 kcal/mol. While individually weak, the large number of non- covalent interactions m a polypeptide having more than several am o acids add up to a large thermodynamic force favoring folding.
  • Protein folding parameters include, among others, those relating to relative hydrophobicity, i . e . , preference for the hydrophobic environment of a non-polar solvent . See Textbook of Biochemistry with Clinical Correlations, 3 rd Ed., ed. Devlin, T.M., Wiley-Liss, p. 30 (1992)). Hydrophobic interactions are believed to occur not because of attractive forces between non-polar groups, but from interactions between such groups and the water m which they are, cr otherwise would Pe, dissolved.
  • the solvation shell (a nighly ordered, and therefore ther odynamically disfavored, arrangement of water molecules around a non-polar group) around a single residue is reduced when another non-polar residue becomes positioned nearoy during folding, releasing water m the solvation shell into the bulk solvent and thereby increasing the entropy of water solvent. It is estimated that approximately one-third of the ordered water molecules m an unfolded protein' s solvation shell are lost into the bulk solvent upon formation of a secondary structure, and that about another one-third of original solvation water molecules are lost when a protein having a secondary structure folds into its tertiary structure.
  • Ammo acid residues preferring hydrophobic environments tend to be "buried," i.e., those found at least about 95% of the time withm the interior of a folded protein, although positioning on the exterior surface of a globular protein can occur by placing the more polar components of the ammo acid near the exterior surface.
  • the clustering of two or more non- polar side chains on the exterior surface are generally associated with a biological function, e.g., a substrate or ligand binding site.
  • Polar ammo acids are typically found on the exterior surface of globular proteins, where water stabilizes the residue's polarity.
  • Positioning of an ammo acid having a charged side chain m a globular protein' s interior typically correlates with a structural or functional role for that residue with respect to biological function of the protein.
  • a hydrogen bond (having bonding energies between about 1 to about 7 kcal/mol) is formed through the sharing of a hydrogen atom between two electronegative atoms, to one of which the hydrogen is covalently ponded (the hydrogen bond "donor") .
  • Hydrogen bond strength depends primarily on the distance between the hydrogen bond donor and acceptor atoms, with high bond energies occurring when the donor and acceptor atoms are from about 2.7 A to about 3.1 A apart.
  • Also contributing to hydrogen bond strength is bond geometry. Bonds having high energies typically have the donor, hydrogen, and acceptors disposed m a colmear fashion. The dielectric constant of the medium surrounding the bond can also influence bond strength.
  • Van der Waals forces which involve both attractive and repulsive forces that depend on the distances between atoms. Attraction is believed to occur through induction of a complementary dipole in the electron density of adjacent atoms when electron orbitals approach at close distances. The repulsive component, also called steric hindrance, occurs at closer distances when neighboring atoms' electron orbitals begin to overlap. With regard to these forces, the most favorable interaction occurs at the van der Waals distance, which is the sum of the van der Waals radii for the two atoms. Van der Waals forces
  • Waals distances range from about 2.8 A to about 4.1 A. While individual van der Waals interactions usually have an energy less than 1 kcal/mol, the sum of these energies for even a protein of modest size is significant, and thus these interactions significantly impact protein folding and stability, and, ultimately, function.
  • folding begins through short range non-covalent interactions between several adjacent (as determined by primary structure) ammo acid side chain groups and the polypeptide chain to which they are covalently linked. These interactions initiate folding of small regions of secondary structure, as certain R groups have a propensity to form ⁇ -helices, ⁇ structures, and sharp, turns or bends m the polypeptide backbone. Medium and long range interactions between more distant regions of the protein then come into play as these distant regions become more proximate as the protein folds.
  • ⁇ As discussed herein, functional site descriptors according to the invention can be developed for any biological function of a protein. Such functions include, but are not limited to, catalysis, ligand binding, and protem-protem interaction Typically, a protein biological function is carried our, or provided for, by one or more specific sites or regions of a protein, and the functionality of such sites are dictated by the three-dimensional structure of the protein.
  • Functional sites of a particular interest m the context of this invention include enzyme active sites, ligand binding sites, and protem-protem interaction sites.
  • Preferred ligand binding sites include substrate binding sites, antigen binding sites, and co-factor binding sites.
  • the instant invention embodies functional site descriptors prepared for the enzymatic activities of the enzyme classes listed above, as well as for other protein activities and functions, whether now known or later discovered.
  • Protein structures can be determined by a variety of experimental or computational methods, several of which are described below.
  • Protein structure can be assessed experimentally by any method capable of producing at least low resolution structures. Such methods currently include x-ray crystallography and nuclear magnetic resonance (NMR) spectroscopy. Structures elucidated by these methods are of varying quality. For the production of functional descriptors according to the invention described herein, high resolution or high quality structures are desirable. Such high quality structures are can also easily be probed for functional sites using the functional descriptors described herein. However, the inventors have discovered that functional site descriptors can also be applied to predict the function m lower quality structures elucidated experimentally, such as low-resolution x-ray crystal structures, in addition to models produced purely by computational methods.
  • functional site descriptors can also be applied to predict the function m lower quality structures elucidated experimentally, such as low-resolution x-ray crystal structures, in addition to models produced purely by computational methods.
  • X-ray crystallography is one method for protein structural evaluation, and is based on the diffraction of X- ray radiation of a characteristic wavelength by electron clouds surrounding the atomic nuclei m the crystal.
  • X-ray crystallography uses crystals of purified biomolecules (but these frequently include solvent components, co- factors, substrates, or other ligands) to determine near atomic resolution of the atoms making up the particular biomolecule.
  • Techniques for crystal growth are known m the art , and typically vary from biomolecule to biomolecule. Automated crystal growth techniques are also known.
  • Small molecules i.e., those having a molecular weight of less than about 2,000 daltons (D)
  • D daltons
  • proteins are typically much larger (typically having molecular weights of 30,000-200,000 D) , and when packaged into crystal lattice points, leave much larger gaps for inclusion of other molecules m the crystal.
  • protein crystals typically contain 40-60% solvent.
  • Dynamic disorder can be reduced or eliminated by lowering the environmental temperature of the crystal during X-ray bombardment. Remaining static disorder is due to one or more rigid static molecular conformations.
  • Detection of diffracted radiation enables the use of mathematical equations (e.g., Fourier synthesis) to generate three-dimensional electron density maps of the diffracted biomolecule. Multiple diffractions are required to make such determinations, with the number of reflections correlating positively with the resolution desired. Low numbers of reflections typically do not provide the requisite information to determine atomic positioning, although the position of a polypeptide chain m individual protein molecules can often be fitted to the electron density map. Structures resulting from these types of crystallographic data are often termed low resolution structures.
  • mathematical equations e.g., Fourier synthesis
  • a protein's ammo acid sequence for example, the primary structure of a protein solved by deducing the ammo acid sequence encoded by a nucleic acid (e.g., a cDNA sequence) encoding the protein
  • a nucleic acid e.g., a cDNA sequence
  • Nuclear magnetic resonance (NMR) currently enables determination of the solution conformation (rather than crystal structure) of biomolecules.
  • small molecules for example proteins of less that about 100-150 ammo acids, are amenable to these techniques.
  • recent advances have lead to the experimental elucidation of the solution structures of larger proteins, using such techniques as isotopic labeling.
  • the advantage of NMR spectroscopy over x-ray crystallography is that the structure is determined m solution, rather than a crystal lattice, where lattice neighbor interactions can alter the protein structure.
  • the disadvantage of NMR spectroscopy is that the NMR structure is not as detailed or as accurate as a crystal structure.
  • biomolecule structures determined by NMR spectroscopy are of moderate resolution compared relative to those determined by crystallography.
  • NMR uses radio frequency radiation to examine the environment of magnetic atomic nuclei m a homogeneous magnetic field pulsed with a specific radio frequency. These pulses perturb the nuclear magnetization of those atoms with nuclei of nonzero spin. Transient time domain signals are detected as the system returns to equilibrium. Fourier transformation of the transient signal into a frequency domain yields a one-dimensional NMR spectrum. Peaks in these spectra represent chemical shifts of the various active nuclei. The chemical shift of an atom is determined by its local electronic environment. Two-dimensional NMR experiments can provide information about the proximity of various atoms in the structure and m three dimensional space.
  • Protein structures can be determined by performing a number of two- (and sometimes 3- or 4-) dimensional NMR experiments and using the resulting information as constraints m a series of protein folding simulations. See Protein NMR Spectroscopy, Principles and Practice, J. Cavanagh, et al . ,
  • experimentally solved biomolecule structures are used to generate and validate functional site descriptors produced m accordance with the instant methods .
  • new functional site descriptors can be created.
  • functional site descriptors produced m accordance with tne invention prior to the solution of such structures can be modified, if necessary. Therefore, certain embodiments of the invention concern the periodic validation, potential modification, and novel creation of one or more of functional site descriptors. These updates are performed through comparison to newly described biomolecules having specified biological function (s), for which three dimensional structures have been elucidated, preferably experimentally (as opposed to prediction) to high resolution.
  • Homology Modeling Techniques Some methods for predicting structures of proteins involve homology modeling. Homology modeling is applied to ammo acid sequences that are evolutionarily related, i . e . , they are homologous, such that their residue sequences can be aligned with some confidence. In one example of this method, the sequence of a protein whose structure has not been experimentally determined is aligned to the sequence of a protein whose structure is known using one of the standard sequence alignment algorithms (Altschul, et al . (1990), J. Mol . Biol . , vol . 215:403-410; Needleman and Wunsch (1970), J. Mol . Biol . , vol .
  • Homology modeling algorithms for example, Homology (Molecular Simulations, Inc.), build the sequence of the protein whose structure is not known onto the structure of the known protein. The result is a predicted model for the sequence whose structure has not been experimentally determined. Such a predicted structure is termed a "homology model" .
  • Threading Algorithms in an inverse folding approach to protein structure prediction one "threads" a probe ammo acid sequence through different template structures and attempts to find the most compatible structure for a given sequence.
  • sequence-to-structure alignments are performed by a "local -glooal” version of the Smith-Waterman dynamic programming algorithm (Waterman, 1995) .
  • alignments are ranked by one or more, preferably three, different scoring methods. In a three method approach (Jaroszewski et al . , 1997), the first scoring method can be based on a sequence-sequence type of scoring.
  • the Gonnet mutation matrix can be used to optimize gap penalties, as described by Vogt and Argos (Vogt et al . , 1995) .
  • the second method can use a sequence- structure scoring method based on the pseudo-energy from the probe sequence "mounted" the structural environment in the template structure.
  • the pseudo-energy term reflects the statistical propensity of successive ammo acid pairs (from the probe sequence) to be found m particular secondary structures withm the template structure.
  • the third scoring method can concern structure-structure comparisons, whereby information from the known template structure (s) is (are) compared to the predicted secondary structure of the probe sequence.
  • a particularly preferred secondary structure prediction scheme uses a nearest neighbor algorithm.
  • the statistical significance of the each score is preferably determined by fitting the distribution of scores to an extreme value distribution, and the raw score is compared to the chance of obtaining the same score when comparing two unrelated sequences (Jaroszewski et al . , 1997).
  • a representative example of automated modeling tools include Modeller4 (Tripos Associates, St. Louis) Such tools preferably produce all non-hydrogen atom coordinate files for the three-dimensional model built from the sequence-to- structure alignment provided by the threading algorithm.
  • the final predicted structure is only as good as the sequence alignment produced by the threading algorithm, and local misalignments may occur m threading predictions and sequence alignments.
  • the problem can be overcome m at least some cases by allowing for small errors m the alignments and by using not ust the threading prediction with the highest score (i.e., the optimum alignment), but a number of top ranking, alternative threadmg-based structure predictions for the same sequence.
  • Use of the conservation profile analysis further increases confidence m threadmg-based structure predictions by seeking consistency m the sequence alignments; thereby reducing the chances of accidental matches.
  • sequence-to-structure alignments are produced for each sequence. Each of these alignments is scanned for the occurrence and conservation of the known functional site residues identified for the functional descriptor (see Figure 8) . If the functional site residues are conserved, a three dimensional structural model is produced from the sequence-to-structure alignment using a modeling tool, such as Modeller4 (Tripos Associates, St. Louis) . The functional descriptor is then used to probe each model to determine the presence or absence of the functional
  • Ah Ini tio Structure Modeling Another computational approach to protein structure elucidation involves ah ini tio prediction. Such procedures generally have two parts: 1) parameter derivation using information extracted from multiple sequence alignment; and 2) structure assembly (or "folding") and refinement. As those the art will appreciate, any conventional or later-developed ah ini tio biomolecule structure prediction algorithm can be used connection with this aspect of the invention.
  • the "MONSSTER" (Modeling Of New Structures from Secondary and Tertiary Restraints) ah ini tio folding algorithm is used to produce inexact models of protein structures.
  • the MONSSTER algorithm uses a high coordination lattice-based ⁇ -carbon representation for the folding of proteins (Skolnick et al . , 1997) and is modified to incorporate the expected accuracy and precision of the predicted tertiary structures
  • a number of independent simulated annealing simulations from a fully extended initial conformation should be carried out (assembly runs) for each biomolecule, particularly when the biomolecule is a protein. Structures generated by these assembly runs are then clustered, and an example low energy structure from each cluster is subjected to a low temperature, isothermal refinement.
  • the predicted fold preferably is that of lowest average energy, although m some embodiments, several of the next lowest energy structures are also examined.
  • inexact protein structure models generated by ah ini tio prediction methods can be utilized for functional analysis using functional site descriptors.
  • functional site descriptors are tested and validated on a series of correctly and incorrectly folded structures produced during the assembly and isothermal runs for various proteins.
  • functional site descriptors can also be used for purposes other than predicting biomolecule function.
  • functional site descriptors can be used to validate or confirm the accuracy of biomolecule structures predicted using the methods described herein or other structure prediction methods. For example, if a biomolecule is predicted to have a particular structure, successful application of one or more functional site descriptors thereto to identify one or more functional sites m the biomolecule validates the predicted fold. Such validation can be further confirmed by experimental demonstration of the particular biological function by the biomolecule. Alternatively, further confirmation of a correct structure prediction by such methods can be derived by the functional site descriptor-mediated identification of two or more functional sites (for example, the substrate binding site and a cofactor binding site) on a biomolecule.
  • the inventors have introduced the concept of a functional site descriptor to help identify molecular functions, as well as techniques for creating functional site descriptors for various functional sites of known molecules, or of known molecular functions.
  • the inventors have further conceived and reduced to practice techniques for utilizing one or more known functional site descriptors to predict functions in a given molecular model.
  • FIG. 2 is an operational flow diagram generally illustrating a process for creating a functional site descriptor for a given molecular function, and using the functional site descriptor to determine whether a given molecular model should exhibit the functional properties associated with the functional site descriptor according to one embodiment of the invention.
  • m step 122 a library of known molecular structures is compiled. To a certain extent, there are libraries already existence and available to many scientists m which known molecular structures for proteins and other biomolecules are catalogued. In accordance with one aspect of the invention, such known and existing libraries can be utilized. These libraries can be further augmented or supplemented as additional molecular structures are identified and characterized.
  • the embodiment described m the context of FIG. 2 concerns a biomolecule that is a protein, particularly an enzyme the catalytic activity of which (at least with regard to the representative biomolecular function herein described) is defined by a known active site.
  • a protein structure library is utilized to build a library of known functional sites.
  • These functional sites are comprised of a group of two or more residues m a molecular structure (e.g., folded protein m the case of proteins) that are known to perform or correlate with a given function or activity for that biomolecule (here, a protein) .
  • a library or database of known active sites (or other functional sites) and their descriptions can be created and maintained. Such descriptions preferably concern the geometric, or three-dimensional, relationships between two or more elements of these functional sites.
  • such relationships are expressed as interatomic distance ranges, and may or may not include information regarding bond angles (or bond angle ranges), between adjacent atoms.
  • descriptions include the identity of the atom, or molecule, at a particular location.
  • representations of ammo acid residues used to describe a functional site descriptor for a particular catalytic function will include the identity of the residue at the particular position.
  • identity of the residue of a particular position m the active site (or other functional site) may not be universally conserved across all members of the particular enzyme (or other biomolecule) class. Thus, m such cases it is preferred that the identity parameter be relaxed and include residues known to reside at that position.
  • ammo acid residue m cases where the biomolecule is a protein
  • the ⁇ -carbon of the residues selected as comprising the active site descriptor may be used.
  • the mean center of mass of the side chain of the particular residues may be selected.
  • other representations or combinations of elements can also be used, and are left to the discretion of the artisan.
  • a group of one or more active sites from the known active site database or library are used to construct a functional site descriptor.
  • the functional site descriptor one embodiment is a somewhat generalized model or description of the active sites or functional sites associated with a particular molecular function.
  • the functional site descriptor is general enough to describe or to help identify functional sites having a given function m numerous different and alternative molecules, without being so broad or generic as to identify functional sites having different functions.
  • the manner m which functional site descriptors can be created according to one or more embodiments are described m detail below. Furthermore, it is preferable that functional site descriptors can be created for numerous molecular functions.
  • a molecular structure of a new molecule is compared to one or more functional site descriptors to predict the function of the molecule.
  • the functional site descriptor is broad enough to generally identify the function m all potential manifestations the functional site can have for various molecular structures.
  • it is preferred that the functional site descriptor is not so broad as to encompass alternative functions.
  • a positive comparison between the functional site descriptor and a site in the molecular structure indicate that the molecule under study exhibits the biological function associated with the functional site descriptor.
  • the functional site descriptor can be used to identify a given function m a molecular structure. Techniques for using a functional site descriptor to predict a biological function m a biomolecule, particularly those whose structure is known by molecular modeling, are also described m detail below.
  • FIG. 3 is a functional block diagram generally illustrating functionality that can be associated with the various processes and techniques for generating functional site descriptors and using functional site descriptors to predict molecular function according to various embodiments of the invention.
  • FIG. 3 is a functional block diagram intended only to exemplify functionality of one or more embodiments described herein.
  • FIG. 3 is not intended to, and should not be construed as, implying a particular physical, logical, or structural architecture for implementing a computer-based system to carry out the functionality described herein.
  • the process begins with a collection or library of known molecular structures as indicated by block 142.
  • these known structures can come from generally available scientific libraries cataloging known molecular structures and/or from the scientific literature.
  • the library will contain at least one known structure.
  • the library will contain two or more (e.g. 3, 4, 5, 6, 7, 8, 9) structures known to possess or be responsible for the particular biological function to be described or represented by the particular functional site descriptor. It will also be appreciated that more than one functional site descriptor can be developed accordance herewith with respect to a particular biological function.
  • a subset of known functional sites 144 here, for purposes of illustration only, active sites
  • the known functional sites are a collection of two or more ammo acid residues that define the particular activity.
  • other atoms and molecules can be used to describe other types of functional sites.
  • known data set 144 is a computer-based data set wherein the information is stored electronically and accessible by a computing system.
  • the data describing the functional sites is a three-dimensional data representation indicating the residues that make up the active site and a spatial relationship between those residues. Specifics regarding how the data are stored and compiled are somewhat secondary, as will be readily understood by one of ordinary skill m the art after reading this description.
  • Processing system 146 utilizes the data representations of the functional s ⁇ te(s) to create one or more functional site descriptors 148.
  • one functional site descriptor is provided for each function associated with a group of functional sites.
  • Processing system 146 can be implemented, for example, as a general purpose processing system. After reading this description, and the further description of the functionality performed by this processing system, one of ordinary skill m the art will understand how to specify and implement a processing system to carry out the described functionality. In one embodiment, processing system
  • the processor-based system 146 is implemented as a Pent ⁇ um ® -based processing system with the appropriate user interfaces and peripheral devices. To predict the functionality of a given molecular structure, the processor-based system 146 compares one or more functional site descriptors 148 with the subject molecular structure 150 to determine if there is a match. If there is a matcn, this provides an indication that the subject molecular structure 150 has the biological function identified as being associated with the matching functional site descriptor 148. In preferred embodiments of such automated implementation of one or more functional site descriptors, the computer-band system will identify which biological function (s) is (are) possessed by the biomolecule (s) under study. The particular output of such results can be implemented in any fashion desired by the user.
  • FIG. 4 is an operational flow diagram illustrating a process for building a functional site descriptor for a given function according to one embodiment of the invention.
  • a group of preferably two or more functional sites are selected from a library of known functional sites 144.
  • all of the functional sites selected are tnose functional sites that are known to be associated with a particular biological function of interest.
  • the geometries of the selected functional sites are determined.
  • the atoms or molecules e.g., ammo acid residues m the case of proteins, particularly with regard to active sites
  • this determination can be made m advance and stored a database, such as a database of known functional sites 144.
  • a database such as a database of known functional sites 144.
  • one or more geometric parameters of the selected functional sites is modified, or "relaxed.” More specifically, m certain preferred embodiments, a delta, or range of possible geometries (for example, distances between the ⁇ -carbon atoms of ammo acid residues) , is applied to a parameter of one or more selected functional site geometries.
  • the modified functional site geometry is compared to the a functional site geometry data set to determine whether this broadened, or relaxed, geometry still compares favorably with the data set of known geometries m its ability to specifically identify biomolecules having the desired biological function.
  • one or more geometric parameters can be further modified or broadened until it is broadened so much that the comparison is no longer favorable because biomolecules known not to possess the particular biological function are identified as having the function. Th s process is illustrated by step 220 and flow line 222.
  • FIG. 5 is an operational flow diagram illustrating a method for creating a functional site descriptor according to one embodiment of the invention.
  • a group of preferably two or more functional site geometries is selected from the known library of functional sites .
  • the functional site geometries are described by N residues and the group of geometries is selected from among structures that are known to have the desired function.
  • three residues are preferably utilized; however, after reading this description it will become apparent to one of ordinary skill m the art how different numbers of residues can be utilized depending on the goals of the system implementation.
  • FIG. 6, which comprises FIGS. 6A, 6B, and 6C, is a diagram illustrating residues of a hypothetical enzyme active site descriptor, adjacent residues, and geometries among the residues according to preferred embodiments of the invention.
  • the active site descriptor is illustrated as being comprised of three residues identified for purposes of illustration only as 3 , 84, and 142, which numbers correspond to ammo acid position m the hypothetical protein depicted m this figure. Amino acid identity at each of those positions has not been provided in this example.
  • residue 84 is identified as being the primary residue.
  • the primary, secondary, tertiary, ... n-ary residues are defined and identified. That is, each of the N residues that make up the functional site descriptor are identified as being primary, secondary, and so on.
  • the selection of which residue is identified as primary can be made somewhat arbitrarily; however, it is preferred that a residue of importance is selected as the primary residue.
  • a biochemist can typically predict, for example, a residue that would go through a covalent change during catalysis. If so, it is preferred that this residue be identified as the primary residue. If there is more than one residue that is known or predicted to go through a covalent change during catalysis, either of these residues is preferably selected as the primary residue.
  • two descriptors each naming a different primary residue (but otherwise comprising the same parameters) , can be generated.
  • additional residues adjacent to the primary, secondary, tertiary, etc. residues that make up the functional site descriptor are selected, as described in step 268.
  • an "adjacent" residue refers to a residue that is either x+1 or x-1 the primary ammo acid sequence if the protein, where x is the ammo acid position of the primary, secondary, etc. residue, as the case may be.
  • FIG. 6b illustrates residues that are immediately adjacent to the residues making up the active site descriptor. Specifically, m the example illustrated in FIG. 6B, residue 84 has adjacent residues 83 and 85. Residue 3 has adjacent residues 2 and 4. Similarly, residue 142 has adjacent residues 141 and 143.
  • m step 270 relative geometries among the residues and the adjacent residues are defined. That is, the spatial relationship between or among the residues is determined and defined. Preferably, the relationship among one or more adjacent residues m relation to other primary, secondary, etc. and/or adjacent residues is also utilized m determining relevant geometries. Where only two or three residues are utilized, and adjacent residues are not utilized, a two-dimensional spatial representation is sufficient from a mathematical perspective. However, as will become more apparent after reading the detailed description below, such a two-dimensional data set may not be sufficient to uniquely define a desired biological function.
  • the data set used to defined the relevant geometries is a three- dimensional data set sufficient to define a three-dimensional relationship among the various residues comprising the functional site descriptor, and preferably includes at least some spatial information with regard to one or more adjacent residues .
  • a nine-distance data set tnat describes nine distances among the various residues and adjacent residues is utilized.
  • FIG. 6C s a diagram illustrating an example of these nine distances for the example residues illustrated m FIG. 6B .
  • FIG. 6C illustrates residue pairings for purposes of determining distance geometries between ⁇ -carbon atoms of ammo acid residues m a hypothetical active site.
  • the residue pairs of this "9-d ⁇ stance" set of geometric constraints depicted are: 84-142, 84-3, 3-142, 83-3, 85-3, 141-84, 143-84 2-84, and 4-84.
  • m the particular circumstance, more or fewer geometric constraints can also be utilized.
  • a data set defining the residues of a functional site can be defined with relative specificity. In fact, m many instances, interatomic distances of various biomolecules are known to accuracies on the angstrom level .
  • m creating a functional site descriptor is to create a descriptor that is broad enough to allow it to match or fall withm the active sites for the same function for many different molecules. That is, for each molecule that performs a particular function, there is functional site geometry for that function. However, exact duplication of spatial, geometric, and other parameters is not necessarily required for duplicate biological function. Therefore, m step 272 of FIG. 5, the geometry of a determined active site descriptor is broadened.
  • one or more geometric or other parameters of a functional site descriptor is broadened by adding a delta, which, the hypothetical under discussion, is preferably a plus or minus uncertainty level, or range, m the distances between the residues selected to comprise the functional site descriptor.
  • a delta which, the hypothetical under discussion, is preferably a plus or minus uncertainty level, or range, m the distances between the residues selected to comprise the functional site descriptor.
  • this new site descriptor is defined, m part, by a plurality of distances, wherein each distance has associated therewith a level of uncertainty.
  • this functional site descriptor having relaxed geometric parameters is compared with one or more functional sites m the data set of known functional sites to determine if the relaxed descriptor accurately identifies all biomolecules m the data set known to have the particular biological function correlated with the descriptor.
  • the functional site descriptor is also compared with biomolecules known not to exhibit the functionality associated with that functional site. That is, the descriptor is also compared to structures or molecules known not to nave that function. If the functional site descriptor known to have a specified function matches or compares favorably to only biomolecules known to have the particular function, and not to biomolecules known not to have that function, the geometric parameters (or other parameters comprising the functional site descriptor) can be expanded (or relaxed) further.
  • the delta associated with one or more interatomic distances, and preferably all of such distances m a given functional site descriptor can be expanded by a specified amount, i.e., by a "multiplier.”
  • tne functional site descriptor can be applied again to the data set to determine whether it matches a sufficient number of the existing active sites known to have this function, without encompassing structures that are known not to have this function. This is illustrated by steps 276 and 278 and flow line 280 m FIG. 5.
  • steps 276 and 278 and flow line 280 m FIG. 5 geometries are adjusted to define a functional site descriptor that is broad enough to encompass active sites exhibiting the same functionality yet not so broad as to encompass active sites that do not have the desired functionality.
  • FIG. 7 is an operational flow diagram illustrating a process for adjusting one or more geometric distance parameters of a functional site descriptor according to the invention.
  • m step 312 one or more, and preferably all, of the distances between residues of the functional site descriptor are averaged by comparing such distances (or mean distances) m
  • Known structures having the particular biological function correlated with the particular functional site descriptor are known structures having the particular biological function correlated with the particular functional site descriptor.
  • a standard deviation for each distance parameter is tnen calculated.
  • a multiplier is applied to the standard deviation to further relax the parameter geometry.
  • each standard deviation is multiplied by the selected multiplier, and the distance between a residue pair is defined as the average distance between that residue pair across the group, plus or minus the standard deviation multiplied by the multiplier.
  • such a relaxed functional site descriptor can then be applied to the data set to determine how many known active sites can be positively identified by the particular functional site descriptor.
  • this comparison can be performed m an integrated fashion, and one or more of the parameters of the descriptor can be broadened on subsequent iterations to encompass additional biomolecules m the data set known to have the desired biological function.
  • the broadening can be accomplished by increasing the multiplier by which standard deviations are multiplied.
  • different multipliers can be applied to different distance parameters, as those in the art will appreciate.
  • the multiplier chosen is the same for each distance m the data set.
  • FIG. 8 is a diagram illustrating an example data set for geometric constraints of a "9 distance" functional site descriptor developed m accordance with instant process. This descriptor describes the active site of phospholipase A2.
  • the data set m FIG. 8 is illustrated m tabular form for ease of description.
  • the rows m FIG. 8 correspond to the distance parameters (written as average distances plus or minus a standard deviation multiplied by a multiplier (here, 2.0)) of the functional site descriptor (here, for the active site for enzymes having phospholipase activity) .
  • a multiplier here, 2.0
  • the columns m FIG. 8 provide the pertinent data for each residue pair distance. In the example illustrated m FIG. 8, these include the average distances for the residue pairs, the standard deviation of these distances among the data set, and the multiplier 326 used to broaden the geometry to the desired breadth.
  • the functional site descriptor is created with a process that includes steps of training the data set to enhance the results.
  • FIG. 9 is an operational flow diagram illustrating a process for creating a functional site descriptor including techniques for training the data set according to one embodiment of the invention.
  • a step 342 a functional site geometry having a known function is selected.
  • the primary, secondary, tertiary and so on residues are identified withm that functional site geometry. This is illustrated by step 344.
  • the identity (les) for each residue is (are) chosen. Preferably, these choices are made based on an analysis of existing structures or other information relating to the particular biological function under consideration .
  • step 350 the distance geometries are calculated. In preferred embodiments, this is accomplisned as described above, whereby the distances between selected residue pairs are computed, averaged for the data set, and adjusted based on the standard deviation multiplied by some multiplier. Multipliers typically are the range of about 0.1 to 100 for distance geometries, with multipliers of 1.0, 1.5, and 2.0 being particularly preferred.
  • the functional site descriptor is screened against a relatively large testing set.
  • the testing set includes molecules of structures known to contain the active site of interest, as well as molecules or structures that are known to not contain that active site.
  • the screening is done with gradually increasing the relaxation of one or more of the parameters of the descriptor (e.g., by increasing a distance geometry multiplier) m an iterative fashion.
  • m a step 354
  • one molecule is removed from the test set and put into a training set upon each iteration.
  • the iterative process continues until the functional site descriptor comprises parameters that fit well withm the data set. This training process is referred to m the computer arts as jack-knife training.
  • FIG. 10 is an operational flow diagram illustrating a process for utilizing a functional site descriptor to determine whether a nucleotide and/or ammo acid sequence encodes a particular biological function.
  • the sequence for testing is identified. This sequence can be a new sequence tnat has just been identified, or an existing sequence. If tne sequence is a nucleotide sequence, the ammo acid sequence s deduced using standard algorithms before proceeding to tne next step.
  • the biomolecule encoded thereby e.g., a protein, enzyme, ribozyme, etc.
  • FIG. 10 assumes that the sequence encodes a naturally occurring protein, for example, an enzyme.
  • a three-dimensional model of the enzyme having that primary sequence is created, for example, by application of a threading, ah ini tio, or other protein folding algorithm.
  • a threading, ah ini tio, or other protein folding algorithm any algorithm useful m generating three-dimensional structural models can be used, e.g., homology modeling, threading or ah ini tio folding algorithms .
  • a residue combination that includes the residues associated with the functional site geometry (or of the functional site descriptor) of the subject function are identified from the three-dimensional folded sequence. For example, if the active site descriptor for the functional site under consideration includes three residues, a Cys, Cys, and a Pro residue, the combinations of the cys, cys, and pro residues are identified withm the folded sequence. Initial filtering can be performed to screen out combinations that obviously are not of the appropriate geometry to fall withm the geometry associated with the active site descriptor.
  • the identified residue combination is compared with the functional site descriptor to determine whether there is a match.
  • the subject folded sequence can be predicted to have the desired function as illustrated by steps 432 and 434. Additional residue combinations can be compared with the functional site descriptor to determine whether these additional combinations match the functional site descriptor. This is illustrated by step 436. Once the residue combinations have been identified for the functional site descriptor, the process can be repeated for other functional site descriptors and other combinations of residues that may comprise other functional sites m that protein. This is illustrated by step 438.
  • the various embodiments, aspects, and features of the invention described above may be implemented using hardware, software, or a combination thereof and may be implemented using a computing system having one or more processors. In fact, m one embodiment, these elements are implemented using a processor-based system capable of carrying out the functionality described with respect thereto.
  • An example processor-based system 502 is shown m FIG. 11.
  • the computer system 502 includes one or more processors, such as processor 504.
  • the processor 504 is connected to a communication bus 506.
  • Various software embodiments are described in terms of this example computer system.
  • the embodiments, features and functionality of the invention as described above are not dependent on a particular computer system or processor architecture or on a particular operating system. In fact, given the instant description, it will be apparent to a person of ordinary skill in the relevant art how to implement the invention using other computer or processor systems and/or architectures .
  • Processor-based system 502 can include a mam memory 508, preferably random access memory (RAM) , and can also include a secondary memory 510.
  • the secondary memory 510 can include, for example, a hard disk drive 512 and/or a removable storage drive 514, representing a floppy disk drive, a magnetic tape drive, an optical disk drive, etc.
  • the removable storage drive 514 reads from and/or writes to a removable storage medium 518.
  • Removable storage media 518 represents a floppy disk, magnetic tape, optical disk, etc. which is read by and written to by removable storage drive 514.
  • the removable storage media 518 includes a computer usable storage medium having stored therein computer software and/or data.
  • secondary memory 510 may include ether similar means for allowing computer programs or other instructions to be loaded into computer system 502.
  • Such means can include, for example, a removable storage unit 522 and an interface 520. Examples of such can include a program cartridge and cartridge interface (such as that found in video game devices) , a removable memory chip (such as an EPROM, or PROM) and associated socket, and other removable storage units 522 and interfaces 520 which allow software and data to be transferred from the removable storage unit 518 to computer system 502.
  • Computer system 502 can also include a communications interface 524.
  • Communications interface 524 allows software and data to be transferred between computer system 502 and external devices.
  • Examples of communications interface 524 can include a modem, a network interface (such as, for example, an Ethernet card), a communications port, a PCMCIA slot and card, etc.
  • Software and data transferred via communications interface 524 are m the form of signals which can be electronic, electromagnetic, optical or other signals capable of being received by communications interface 524. These signals are provided to communications interface via a channel 528. This channel 528 carries signals and can be implemented using a wireless medium, wire or cable, fiber optics, or other communications medium.
  • a channel can include a phone line, a cellular phone link, an RF link, a network interface, and other communications channels.
  • computer program medium and “computer usable medium” are used to generally refer to media such as removable storage device 518, a disk capable of installation m disk drive 512, and signals on channel 528.
  • Computer programs also called computer control logic
  • Computer programs can also be received via communications interface 524. Such computer programs, when executed, enable the computer system 502 to perform the features of the present invention as discussed herein.
  • the computer programs when executed, enable the processor 504 to perform the features of the present invention. Accordingly, such computer programs represent controllers of the computer system 502.
  • the software may be stored , or transmitted via, a computer program product and loaded into computer system 502 using removable storage drive 514, hard drive 512 or communications interface 524.
  • the control logic when executed by the processor 504, causes the processor 504 to perform the functions of the invention as described herein.
  • the elements are implemented primarily m hardware using, for example, hardware components such as PALs, application specific integrated circuits (ASICs) or other hardware components.
  • ASICs application specific integrated circuits
  • implementation of a hardware state machine so as to perform the functions described herein will oe apparent to persons skilled m the relevant art(s)
  • elements are implemented using a combination of both hardware and software.
  • the active sites responsible for the disulfide oxidoreductase activity of the glutaredoxm/ thioredoxm family m E. coli and the RNA hydrolytic activity of the TI ribonuclease family are described as functional site descriptors. Then, these descriptors are shown to correctly identify the appropriate active site (for example, that enabling oxidoreductase catalytic activity) m a library of high resolution protein structures produced by X-ray crystallography or NMR spectroscopy, most of which proteins had previously been identified as having other activities, but not oxidoreductase or ribonuclease activity.
  • these functional site descriptors are used to screen for and successfully identify active sites m low-to-moderate resolution structures predicted by ah ini tio folding or threading algorithms. Also described is the prediction of a function of a yeast protein whose three-dimensional structure was predicted using a threading algorithm.
  • Automation of these methods, m conjunction with libraries of functional site descriptors, enables the large scale functional screening of nucleotide or protein sequence databases, first by predicting one or more three-dimensional structures from one or more sequences, then by identifying functional sites, e.g., active site, m the predicted structure (s) .
  • YME3_THIFFE is a predicted 9.0 kD protein m the MOBE 3' region (ORF 8) in Thiohacillus ferrooxidans .
  • ORF 8 the MOBE 3' region in Thiohacillus ferrooxidans .
  • a clone containing this gene complements an E. coli thioredoxin mutant (Rohrer and Rawlmgs, 1992), providing experimental evidence that this protein falls into the glutaredoxm/ hioredoxin family.
  • a BLAST search of a non-redundant sequence database (Genbank CDS translations, PDB, SwissProt, and PIR; http : //www.ncbi .nlm.nih.gov/BLAST/blast-databases .html) using YME3_THIFE as the search sequence identified two significant matches (to a pe ⁇ plasmic hydrogenase from D . vulgaris (PHFL_DESVO) and 0RF-R5 from Anahaena) and two potential matches (one of which is GLRX_METTH, a glutaredoxm-like protein from Me anobacterium t ermoa o rop icum) .
  • GLRX_METTH and YME3_THIFE show conservation of the active site cystemes.
  • YME3_THIFE has weak sequence similarity to known thioredoxins, and experimental evidence supports such an assignment, the sequence is not identified as such by a comparison with the local sequence motifs of the Prosite database because it contains only a portion of either the glutaredoxm or thioredoxin Prosite local sequence motifs.
  • the YME3_THIFE sequence is also not identified as a thioredoxin by Prints, although it classified as a glutaredoxm by the Blocks database .
  • the glutaredoxm/thioredoxm protein family is comprised of small proteins that catalyze thiol-disulfide exchange reactions via a redox-active pair of cystemes m the active site (Yang & Wells, 1991a; Yang & Wells, 1991b) . While glutaredoxms and thioredoxins catalyze similar reactions, they are distinguished by their differential reactivity.
  • Glutaredoxms contain a glutathione binding site, are reduced by glutathione (which is itself reduced by glutathione reductase) , and are essential for the glutathione-dependent synthesis of deoxyribonucleotides by ribonucleotide reductase (Holmgren & Aslund, 1995) .
  • glutathione which is itself reduced by glutathione reductase
  • thioredoxins are reduced directly by the flavoprotem, thioredoxin reductase, and act as more general disulfide reductases (Holmgren & Bjornstedt, 1995).
  • reducing equivalents for both proteins come from NADPH.
  • Protein disulfide isomerases have been found to contain a thioredoxm- like domain and thus also have a similar activity (Kemmmk et al., 1997; Kemmmk et al . , 1995).
  • the active site of the redoxm family contains three invariant residues: two cystemes and a cis-proline. Mutagenesis experiments have shown that the two cystemes separated by two residues are essential for significant protein function. The side chains of these two residues are oxidized and reduced during the reaction (Busnweller et al . , 1992; Yang & Wells, 1991b). These two cystemes are located at the N-terminus of an ⁇ -helix.
  • Another unique feature of the redoxm family is the presence of a cis-prolme located close to the two cystemes folded, three-dimensional structures, but not m the primary structures, i . e . , the one- dimensional representations of the ammo acid sequences.
  • the functional site descriptor for the disulfide oxidoreductase activity (i.e., a biological function) of the glutaredoxm/thioredoxm family was built as follows using the three ammo acids discussed above: the structure of the active site was taken from the three-dimensional structural comparison of bacteriophage T4 glutaredoxm, laaz (Eklund et al . , 1992), human thioredoxin, 4trx (Forman-Kay et al . , 1990), and disulfide bond formation protein, ldsb (Martin et al . , 1993) .
  • the active sites of these three proteins can be aligned the ⁇ -carbon distances of the two cysteme ammo acids and the cis-prolme ammo acid of the functional site descriptor appearing m FIG. 12.
  • the following disulfide oxidoreductase functional site descriptor was thus developed: two cystemes separated by two residues and an ⁇ -carbon distance of 5.5 +/- 0.5 A. These cysteines must be close to a proline residue. Specifically, the ⁇ -carbon distance from Cys(i) to the proline is 8.5 +/- 1.5 A and that from Cys(i+3) to the proline is 6.0 +/- 1.0 A.
  • the distances-only functional site descriptor from section (b) is almost sufficient to uniquely distinguish proteins belonging to the glutaredoxm/thioredoxm family from a data set of 364 non-redundant proteins whose structures have been previously elucidated taken from the Brookhaven database. For this set of 364 proteins, 13 have the sequence signature -C-X-X-C- . Of these, only three, lthx (thioredoxin) , IdsbA (protein disulfide isomerase, chain A) , and lprcM (photosynthetic reaction center, chain M) , have a prolme with the distances specified m the descriptor.
  • the lprc-M site is also eliminated, making the functional site descriptor absolutely specific for the glutaredox /thioredoxm disulfide oxidoreductase function.
  • the allowed variance m the Cys-Pro and Cys-Cys ⁇ -carbon distances was uniformly increased m increments of +/- 0.1 A.
  • llct an iron transport protein
  • cysteines are m a very irregular structure, not m a helix.
  • Cys-Cys-Pro structural motif is located along one face of a transmembrane helix, near the helix's C-termmus.
  • Cys-Cys- Pro are located m another very irregular region.
  • a ferredoxm also a metal-bmdmg protein
  • the cysteines were found m a nonregular structural region, not m a helix.
  • the functional site descriptor includes the conformational parameter that the cysteines be located at the N-termmus of a helix, all false positives can be excluded, even when the functional site descriptor distance constraints are relaxed by +/- 0.3 A.
  • the best inexact model produced by MONSSTER is 5.7 A RMSD between corresponding ⁇ - carbons .
  • sequence of this glutaredoxm exhibits less than 30% sequence identity to any of the three structures used to create the disulfide oxidoreductase functional site descriptor.
  • the disulfide oxidoreductase functional site descriptor was applied to 25 correct structures and 56 incorrect, or misfolded structures generated by MONSSTER for the lego sequence during the isothermal runs.
  • the distances-only functional site descriptor selected all 25 "correct" structures as belonging to the redoxm family and rejected all 56 misfolded structures.
  • Sequence-to-structure alignments were performed by a "local -global" version of the Smith-Waterman dynamic programming algorithm (Waterman, 1995) .
  • the alignments were ranked by three different scoring methods (Jaroszewski et al . , 1997) .
  • the first scoring method was based on sequence- sequence type scoring using the Gonnet mutation matrix to optimize gap penalties, as described by Vogt and Argos (Vogt et al . , 1995) .
  • the second method was a sequenc -structure scoring metnod based on the pseudo-energy from the probe sequence "mounted" m the structural environment m the known structure.
  • the pseudo-energy term reflected the statistical propensity of successive ammo acid pairs (from the probe sequence) to be found m particular secondary structures within the template structure.
  • the third method was a structure-structure scoring method, whereby information from the known template structure was compared to the predicted secondary structure of the probe sequence .
  • the secondary structure prediction scheme for the probe sequence employed the nearest neighbor algorithm (Rychlewski & Godzik, m preparation) .
  • the version used here achieved an average three-state prediction accuracy of 74%.
  • sequences of four proteins lacking significant homology from the S. cerevisiae genome database were tested using the disulfide oxidoreductase functional site descriptor: one protein was predicted to belong to the protein disulfide isomerase family (S67190) , one sequence was identified in the database as a hypothetical thioredoxin (YCX3_YEAST) , and two hypothetical proteins, one having very distant sequence similarity to glutaredoxin from rice (S51382) and the other with very distant sequence similarity (insignificant by the Blast score) to the glutaredoxin from Methanococcus thermoautoformicum (S70116) . S51382 has not been identified as a glutaredoxin or thioredoxin in any sequence or motif database .
  • the threading algorithm (Godzik & Skolnick, 1992) aligned the sequences of all four known proteins to the structure of either lego (E. coli glutaredoxm (Xia et al . , 1992)) or 2trx (E. coli thioredoxin (Katti et al . , 1990)) from a database of 301 non-homologous proteins (Fischer et al . , 1996).
  • the alignment fit was strong, as the sequences were matched to either lego or 2trx (chain A) by all three scoring methods used to assess the significance of the threading results. Models were built based on the sequence-to-structure alignments and were screened with the functional site descriptor.
  • TI Ribonuclease Functional Site Descriptor (f) TI Ribonuclease Functional Site Descriptor.
  • a functional site descriptor was also developed for the active site of the TI ribonucleases, a family of proteins that include a number of ribonucleases such as TI, T2 , U2 , and FI, and the distantly related family of fungal ⁇ botoxms . These proteins are endoribonucleases generally specific for purme, particularly guanme, bases (Steyaert, 1997) .
  • the TI ribonuclease functional site descriptor was implemented in three stages: first, each structure was searched for the residue triad involved m nucleophilic displacement (His-His- Glu) ; second, those structures having the His-His-Glu triad was searched for the residue triad involved m transition state stabilization (Tyr-Hydrophobic-Arg) ; and third, if both triads were found, the relative positions of the two triads was checked based only on the distances between ⁇ -carbons.
  • Ribotoxms are a small family of proteins found m the Asperg llus fungi family. They cleave rRNA, inactivating the ribosome and ultimately killing the cell (Kao & Davies, 1995) . The RNA cleavage is carried out by a mechanism quite similar to that found in the TI ribonucleases (Campos-Olivas et al . , 1996).
  • the three selected ribotoxms, ⁇ -sarcm (RNAS_ASPGI) , clavin (RNCL_ASPCL) , and restrictocm (mitogillm) (RNMG_ASPRE) can be aligned to the TI ribonucleases by multiple sequence alignment algorithms, but the sequence identity between the ribotoxms and the TI ribonucleases is quite low (less than 35% pairwise sequence identity) . Furthermore, a Blast (Altschul et al . , 1990) search of SwissProt (Bairoch & Apweiler, 1996) using the sequence of 9rnt as the search sequence did not yield any of these ribotoxm sequences. The structures of ⁇ -sarcm (Campos-Olivas et al . , 1996) and restrictocm (Yang & Moffat, 1996) have been solved, but neither has yet been released to a public database.
  • This example demonstrates the instant invention' s ability to predict protein function based on the three-dimensional structure of an active site.
  • the method which is amenable to automation, uses a sequence-to-structure-to-function paradigm wherein the protein' s structure is first predicted from its ammo acid sequence, after which the active site of the protein is identified in the predicted model using a functional site descriptor. Function is then assigned based on the particular functional site descriptor that aligns with the predicted structure of the protein.
  • the instant invention has the following advantages (each is discussed m further detail m the following paragraphs) : 1) it is applicable even when the sequence identity between two proteins is not significant; 2) it can be used with proteins having different global folds, but similar active sites and associated function; 3) it distinguishes between proteins with similar folds (topological cousins) and those that belong to a given functional family; and 4) in addition to assigning a given protein to a functional family, the method produces a three-dimensional map or model of the protein's active site.
  • This example describes automated application of the sequence-to-structure-to-function paradigm of the invention to the complete Escherichia coli genome (i.e., all ORFs) to identify proteins having the thiol -disulfide oxidoreductase activity of the glutaredoxm/thioredoxm protein family.
  • protein structures are predicted by a threading algorithm from ammo acid sequences deduced from nucleotide sequence information.
  • the threading algorithm generates a model of each protein' s structure by aligning its primary sequence to the best matching structure m a structural database and extending the sequence analysis well beyond the limits of local sequence identity.
  • the modeled structure is then probed with a functional site descriptor for the active site responsible for the thiol -disulfide oxidoreductase activity of the glutaredox /thioredoxm protein family, namely the functional site descriptor for oxidoreductase activity described m Example 1.
  • a functional site descriptor for the active site responsible for the thiol -disulfide oxidoreductase activity of the glutaredox /thioredoxm protein family namely the functional site descriptor for oxidoreductase activity described m Example 1.
  • E. coli is a very widespread, medically important organism that colonizes m the lower gut of animals. It is also responsible for a variety of infections and diseases. Furthermore, E . coli is a preferred organism for genetic, biochemical, and molecular biology studies and has been very well studied. Because of its importance m so many realms, it was one of the earliest candidates for genome sequencing (Neidhardt et al . , 1996). The complete 4 , 639 , 221-base pair sequence of the genome of E. coli K-12, strain MG1655, has recently been published (Blattner et al . , 1997), and nearly 40% of the ORFs encoded thereby were previously uncharacterized.
  • Thrd/FSD Alignment of E. coli ORF to the sequences of lego, ldsb (chain A) , or 2trx (chain A) using a threading algorithm (Jaroszewski et al . , 1998), followed by analysis of the resulting sequence-sequence alignment for the active site residues specified by the functional site descriptor for the thiol-disulfide oxidoreductase activity of the glutaredoxm/thioredoxm family (Fetrow & Skolnick, 1998) . Threading results concern a combination of three different scoring methods, sq, br, and tt, as described by Godzik and coworkers (Jaroszewski et al .
  • Sq is a sequence-sequence type of scoring that uses a Gonnet mutation matrix (Gonnet et al . , 1992) to optimize gap penalties (Vogt et al , 1995)
  • Br is a sequence-structure scoring method that is based on the pseudo-energy from the probe sequence
  • the pseudo-energy term reflects the statistical propensity of successive residue pairs (from the probe sequenced) to be found in a particular secondary structure withm the template structure.
  • Tt is a structure-structure scoring method, whereby information from the known template structure is compared to the predicted secondary structure of the probe sequence . Secondary structure was predicted by a nearest neighbor algorithm (Ryschlewski & Godzik, 1997) . From these scores, statistical significance was calculated based on a maximum value distribution (Jaroszewski et al . , 1998) .
  • Blst/FSD Alignment of each E. coli ORF to the sequences of tne lego (Xia et al . , 1992), ldsb, chain A (Martin et al . , 1993), and 2trx, chain A (Katti et al . , 1990) proteins using the BLAST search protocol (Altschul et al . , 1990), followed by analysis of the resulting sequence-sequence alignment for the active site residues specified by the thiol-disulfide oxidoreductase activity of the glutaredox /thioredoxm family.
  • results reported here are for a combination of the gapped-BLAST protocol and the PSI-BLAST alignment protocols (Altschul et al . , 1997) . All sequences marked are found by both gapped- and PSI-BLAST, except YIEJ_ECOLI, which is found only by gapped-BLAST.
  • Functional Motif Search of each sequence found by either Blast or threading protocols against the local signature databases Prosite (ps; http://expasy.hcuge.ch/sprot/ scnpsitl.html), Prints using the Prosite scoring method (pps; http : //www.biochem.ucl .ac .
  • **Prmts has three patterns for glutaredoxm/thioredoxm activity. This sequence only hits two of these patterns.
  • the Prosite database (Release 13 and updates to Nov. 28, 1997; Bairoch et al . , 1995) identifies nine E. coli sequences that exhibit a glutaredoxm (PS00195) or thioredoxin (PS00194) active site. All nine of these proteins were also identified by threading/functional site descriptor procedure. The latter method also identified glutaredoxm 2, GLR2_EC0LI, which had been previously reported to be an atypical glutaredoxm (Vlamis-Gardikas et al . , 1997), and is inconsistently classified by the local sequence motif databases (Prosite, Prints, and Blocks) . In addition, this protein is not recognized as a glutaredoxm or thioredoxin by BLAST.
  • the threading/functional site descriptor procedure also identified three sequences not identified by Prosite as containing the thiol-disulfide oxidoreductase site: NRDH_ECOLI, NRDG_ECOLI , and B0853.
  • NRDH_ECOLI is a small protein found m the nrdEF operon of E. coli that is involved in electron transport for ribonucleotide reductase (Jordan et al . , 1996). Searching the local signature databases with the NRDH sequence showed that the Prosite and Prints databases using the Prosite scoring function do not classify this sequence as a glutaredoxm.
  • a Prints database search using the Blocks scoring function identified two out of the three motifs, and Blocks identified the CXXC motif as a glutaredoxm motif, but the significance score was not high.
  • the authors noted its similarity to some glutaredoxms (Jordan et al . , 1996) .
  • Recent experimental results have demonstrated that the protein is a functional redoxm (Jordan et al . , 1997) .
  • the methods of this invention clearly identified an active site that was not identified by either Blast or Prosite, but was identified by Prints and Blocks with low significance.
  • NRDG_ECOLI The second sequence, NRDG_ECOLI , has been classified as an anaerobic ribonucleoside-t ⁇ phosphate reductase activating protein and is known to participate m redox reactions (Sun et al . , 1995) .
  • B0853 is an ORF that has been classified as a regulatory protein (see http:// www. genetics wisc.edu/html/orftables/mdex.html). B0853 was not recognized by Prosite, Prints, or Blocks as having a glutaredoxm or thioredoxin active site.
  • the methods of the invention accurately identified the activity of all ten of proteins accurately, as did the BLOCKS scoring method (Henikoff Sc Henikoff, 1991) .
  • NRDG_ECOLI and B0853 were also identified by the methods of the invention as having thiol- disulfide oxidoreductase activity.
  • sequence YBBN_ECOLI yielded a very high significance score to 2trx, chain A, by all three scoring methods .
  • Thread Scores These scores (sq, br, and tt, from left to right) are the significance scores described in the footnotes to Table 1.
  • the sequence SGBH_ECOLI was aligned to IdsbA only by the br scoring method, but not by the sq or tt methods.
  • this sequence also had very high sequence identity to many thioredoxins; however, it was not recognized by the functional site descriptor because the most N-terminal of the active site cysteines was changed to a serine.
  • This protein was a topological cousin, and since it just contains a single cysteine, it is unlikely to exhibit significant disulfide oxidoreductase activity; however, it has been shown that a few of these proteins can still catalyze disulfide isomerization reactions when the second cysteine (but not the first) is replaced with serine (Wunderlich et al . i 1995), though in most cases, the activity is significantly reduced (Bushweller et al .
  • YBBN_ECOLI is such a sequence and has become specialized from a general thiol-disulfide oxidoreductase into a redoxin-like protein that can only catalyze disulfide interchange reactions.
  • oxidoreductase functional site descriptor of Example 1 can successfully identify the active site residues in ten sequences that are known or are proposed to exhibit the glutaredoxin/thioredoxin oxidoreductase activity in the well-studied E. coli genome. Furthermore, the method predicts thiol-disulfide oxidoreductase active sites in two other sequences that have not been identified previously.
  • This example concerns the development and use of a functional site descriptor for the active site of ⁇ / ⁇ hydrolases .
  • These enzymes are of significant medical interest.
  • inhibitors of acetylcholinesterase a member of the family, are used in treatment of Alzheimer's disease, myasthenia gravis, and glaucoma.
  • the active site descriptor described in this example is defined in terms of three dimensional ⁇ -carbon coordinates and residue identities. The goal of these efforts was to identify all proteins encoded in the E. coli genome that have an ⁇ / ⁇ hydrolase fold and exhibit a hydrolase activity.
  • a threading algorithm was used to align 651 E.
  • ORFs and 17 of these ORFs were predicted to encode hydrolase activity through application of the ⁇ / ⁇ hydrolase active site descriptor. The putative active site residues of these proteins were also identified.
  • One of these ORFs, YHET_ECOLI was predicted to encode a protein classified as a member of UPF0017 (an uncharacterized protein family) , which bears all the hallmarks of the ⁇ / ⁇ hydrolase family.
  • the ⁇ / ⁇ hydrolases are known to participate in many physiological processes, and much is known about their structure and biochemistry.
  • the family encompasses a wide range of enzymatic functions. Table 5, below, lists some of the known members of this family.
  • the side chain of the Ser residue is a nucleophilic center and the His side chain acts as a general base and be hydrogen bonded to the carboxylic group of the Asp side chain His and Asp together form a charge relay system.
  • the ammo acids surrounding these residues form an oxyanion hole for stabilizing the enzyme's transition state intermediate, and the Gly residues flanking the Ser position provide the structural flexibility required to perform the catalysis (Ollis, et al (1992) , Protein Eng . 5, 197-211) Except for His, none of the active site residues is known to be absolutely conserved throughout the family Thus, His was selected as the primary residue for purposes of building the ⁇ / ⁇ hydrolase active site descriptor.
  • a search for similar nine ⁇ -carbon arrangements m 3D space was performed on a database that contains 1038 non-homologous structures extracted from the FSSP database (Holm and Sander (1997) , Nucleic Acids Res . 25, 231-234) .
  • any three residues were considered (provided one was His) whose C ⁇ _ C ⁇ distances were less than 12 A from one another.
  • RMSD root mean square deviation
  • the 9-C ⁇ scaffold was specific: all of the proteins that are members of the ⁇ / ⁇ hydrolase fold family have a 9-C ⁇ scaffold with an RMSD of less than 1.0 A, as compared to the lgpl active site. Note that these proteins (listed m Table 6, below) had previously been experimentally determined to have hydrolase activity.
  • site 1 (as listed in Table 6) is near to the oxyanion hole, and is believed to participate in forming hydrogen bonds with the substrate; and the site 3 residue (see the various alternatives listed in Table 6) can be hydrogen-bonded to His to form the charge- relay system in the catalytic triad.
  • sites 1-5 listed in Table 6 are structurally conserved across the entire family, albeit with some variation in several of these positions.
  • the sites shown m Table 6 are critical to catalytic function (Ollis, et al . , supra) .
  • the His at site 5 is not replaceable, but residue at site 2 can be either Ser, Asp, or Cys, and that at site 4 can be an Asp or Glu.
  • a hybrid threading algorithm (Jaroszewski, et al . (1998), Protein Sci . 7, 1431-1440) was applied to the whole set of 4289 open reading frames (ORFs) m the E. coli genome (Blattner, et al . (1997), supra) to predict their structures. Briefly, the algorithm threaded each query sequence though a library of structures using dynamic programming. The structural library used was collected from the FSSP database, which contains 1038 nonredundant structures with less than 30% pairwise sequence identity among them, of which 13 were identified previously as being members of the ⁇ / ⁇ hydrolase fold family (see Table 6) .
  • the threading algorithm output the names of the five most compatible protein structures, as well as the alignments between the query sequence and the sequences of those five most compatible structures. Since three scoring functions were used for each query sequence, 15 sequence-to- structure alignments were obtained for each E. coli protein sequence.
  • any of the 15 structures that aligned to a given query sequence that was a member of the ⁇ / ⁇ hydrolase fold family listed m Table 6 was called a "hit” .
  • the distribution of the threading scores was calculated as the logarithm of the significance scores. See Jaroszewski, et al . (1998), Protein Sci . 7, 1431-1440).
  • the total number of hits was 1003, corresponding to 651 different ORFs, i.e., 651 different ORFs had at least one hit to a member of the ⁇ / ⁇ hydrolase fold family.
  • ORFs with one or none conserved triad residue were predicted either not to have hydrolase activity or not to have an ⁇ / ⁇ hydrolase fold.
  • PID is the sequence ID number in GenBank. 2.
  • name ' ' is the gene name of the E coli protein sequence.
  • N is the number of residues of the E coli protein sequence.
  • m is the number of non- redundant sequences that were found to bear sequence similarity to the E. colt protein sequence. These sequences were used to calculate the conservation profile.
  • database annotation means the database annotation of function in SWISS-PROT (marked by the symbol t) or, if no such annotation exists, the function annotation of a similar sequence found in a PSI-BLAST search "*" indicates those structures predicted by threading not to be ⁇ / ⁇ hydrolases, but with threading scores greater than 10. ⁇ indicates proteins for which existing annotations suggest a function other than hydrolase
  • Table 7(a) lists 16 proteins predicted to have three dimensional structures similar to the ⁇ / ⁇ hydrolase fold family and that function as hydrolases. Among these proteins, it is experimentally known that bioH is carboxylesterase and pldB is lysophospholipase L2 , both of which are within the scope of 0 functions listed above; thus, these are confirmed predictions.
  • the other proteins listed in Table 7 (a) are proteins deduced from the E. coli ORFs. Most of them can be related to sequences that have one of the functions assigned m the search of the ⁇ / ⁇ hydrolase fold family by PSI-BLAST. However, there are a few exceptions.
  • the database annotation column lists spermidme synthase, nitrogen fixation activator, and oxygenase, none of which seem similar to the hydrolase functions in listed in Table 6. While these predictions could be the false positives, it is also possible that these proteins were identified because they are multifunctional .
  • the biological function (s) of the proteins listed in Table 7(b) is (are) less certain.
  • a PSI-BLAST search found some of these proteins to be related to lipases and esterases . However, because none of catalytic triad residues of these proteins were observed as being conserved m a multiple sequence alignment of related proteins, these proteins may function differently from the known members of the ⁇ / ⁇ hydrolase family.
  • the predicative power of the instant invention is strongly illustrated by the deduced E . coli protein YHET.
  • the SWISS -PROT database annotation for this ORF indicates that this protein belongs to the uncharacterized protein family UPF0017, which contains n ne previously uncharacterized proteins homologous to YHET_ECOLI from a wide range of organisms, including humans.
  • Table 8 shows the results obtained from the threading procedure for YHET. Table 8. YHET_ECOLI threading scores.
  • the predicted folds are denoted by PDB code names with the fifth letter denoting the chain labels, "sq”, “br”, and “tt” refer to the types of sconng functions. Structures denoted with an ⁇ n asterisk are in the ⁇ / ⁇ hydrolase fold family. Threading score are expressed as the logarithm of the significance score of the threading alignments.
  • IbroA is a bromoperoxidase (Hecht, et al . (1994), Nat . Struct . Biol . 1, 532-537), but YHET and IbroA only share about 20% sequence identity.
  • YHET is an ⁇ / ⁇ hydrolase.
  • YHET as an ⁇ / ⁇ hydroxylase through the identification of a local sequence motif (a "nucleophilic elbow; see Schrag and Cygler (1997), Methods Enzymol . 284, 85-107; Petersen, et al . (1997), Methods Enzymol . 284, 61-85) characterized by the glycmes at the ⁇ +2 and i-2 positions relative to the nucleophilic Ser in YHET active site.
  • a local sequence motif characterized by the glycmes at the ⁇ +2 and i-2 positions relative to the nucleophilic Ser in YHET active site.
  • the biological function prediction method described in this example successfully identified the proteins belonging to the ⁇ / ⁇ hydrolase fold family encoded in the ⁇ . coli genome.
  • This method has three key components: (1) an active site descriptor for the ⁇ / ⁇ hydrolase fold family; (2) a threading algorithm to predict a structure for a query sequence; and (3) a conservation profile produced by a multiple sequence alignment to the query sequence.
  • this method uses structural information (namely a functional site descriptor superimposed on a structure) , it is more specific than sequence motif-based methods, such as BLOCKS, PRINTS, and PROSITE. It also differs from methods that require precise coordinates of the side chain atoms for active site identification, and thus can be employed m conjunction predicted protein structures. As a result, it is particularly useful in conjunction with genome sequencing projects and efforts to determine the biological function (s) of biomolecules.
  • This example describes the application of the inventors' sequence-to-structure-to function paradigm m a scaled up analysis of the complete genomes of Haemophilus influenzae and Methanococcus jannaschii for proteins exhibiting the thiol/disulfide oxidoreductase activity of the glutaredoxin/thioredoxin family.
  • All deduced ammo acid sequences m both genomes were aligned to a set of structural proteins using a threading algorithm.
  • Structural models were then built automatically and screened using the glutaredoxm/thioredoxm functional site descriptor.
  • This FSD was based on three criteria: 1) two cystemes separated by two ammo acids m the primary sequence; 2) a prol e must be located near the two cysteines m three-dimensional space; and 3) the two cysteines must be at the N-termmus of a helix.
  • the allowed distance between the ⁇ -carbons of the two cysteines was 5.5 +/- 0.5 A.
  • the allowed distances between the more N- termmal and the more C-termmal cysteine and the prolme were 8.5 +/- 1.5 A and 6.5 +/- 1.5 A, respectively.
  • H. influenzae genome Analysis of the H. influenzae genome .
  • seven proteins were found by performing a keyword search of the H. influenzae database (http://www.tigr.org/tdb/mdb/hidb/hidb.html) using the keywords glutaredoxin, thioredoxin, or disulfide isomerase.
  • Organism is name of the organism whose genome was search.
  • Keyword is the word that was used m the keyword search of the TIGR databases (either http://www.tigr.org/tdb/mdb hidb/hidb.html or http://www.tigr.org/tigr_home/tdb/mdb/mjdb/mjdb html) Proteins under the "annotated” subheading were known (or suggested) to belong to this family either by experiment or by sequence alignment analysis. HI1159 is probably maccurately annotated because it has no cysteines m its sequence If the keyword subheadmg is "predicted", then the listed sequences were found by application of the FSD to the model produced from the threading alignment and were not found by a keyword search of the genome database.
  • Sequence ID is the database number used to identify each sequence in the respective genome databases
  • Orgamsm/SeqNum is the organism and the number from the genome databases. Refer to Table 1 for the complete names of these proteins and ORFs. "Annotated" sequences are those that are identified in the TIGR databases as glutaredoxms, thioredoxins, or protein disulfide isomerases (see Table 1), while 'predicted” sequences are those that are not idennfied in the genome databases, but are predicted to have the acnve site by application ot the FFF to the three-dimensional model produced from the threading alignment.
  • X indicates that the sequence was recognized by the searching method used; XI and X2 indicate that three local motifs were used to identify the function, but only one or two, respectively, of the motifs was found in the sequence; "-" indicates that the sequence was not recognized by the method.
  • the last column indicates the putative active site residues identified by the FFF m the threading model.
  • the glutaredoxm pattern (PR00160A) was the sixth highest sconng pattern found for this sequence, with a percent e score of 23.
  • the C4-type steroid receptor was the highest ranking sconng pattern found for this sequence in the Prints database using the Blocks scormg method.
  • One of the other three sequences (HI0846 1 was recognized by Prosite and Blocks; one sequence (HI1159) was only recognized by the Prints database (using the Blocks scoring method) ; the other sequence (HI0428) was not recognized by any of the motif databases (Table 2) . It is unclear whether or not these last three sequences contain the disulfide oxidoreductase activity. HI1159 is probably incorrectly annotated in the H. influenzae genome database because this sequence contains no cysteines, residues necessary of the oxidoreductase reaction mechanism of the glutaredoxin/thioredoxin family, even though it otherwise exhibits some sequence similarity to a thioredoxin-like protein from a cyanobacterium.
  • the FSD identified 11 additional sequences from the H. influenzae genome as containing the disulfide oxidoreductase active site. These sequences and the proposed active site residues are presented in Tables 1 and 2 as predictions. Comparison to the analysis of these sequences by BLAST, the motif databases (Prosite, Prints, and Blocks), and GeneQuiz (43,44) is also presented in Table 2. Most of these 11 sequences are those that align to lego, 2dsb, and 2trx via the threading algorithm with the lowest significance scores.
  • BLAST aligned all three of these sequences to other thio/glutaredox s or thiol/disulf de interchange proteins m the non-redundant sequence database.
  • HI0434 listed m the database as a comF protein, aligned to several competence proteins and hypothetical proteins from other organisms.
  • nrfB formate-dependent nitrite reductase
  • M. Jannaschii genome one protein, MJ0307, was identified as a hypothetical thoredoxm by the keyword search of the TIGR database. This protein exhibits significant sequence similarity to other thioredoxins, and is recognized by the local sequence motif databases Blocks and Prints using the Blocks scoring method, but s not recognize ⁇ oy Prosite or by Prints using the Prosite scoring method.
  • MJ0757, MJ1342, and MJ1552 are hypothetical proteins.
  • the MJ0757 sequence only has significant similarity to other hypothetical proteins.
  • MJ1342 has no significant sequence identity to other proteins, while MJ1552 exhibits significant sequence similarity only to hypothetical proteins from other organisms and to one acyl synthase. The fact that regions of those proteins match the disulfide oxidoreductase FSD provides evidence that these proteins exhibit oxidoreductase activity.
  • MJ0156 is identified as an acetylCoA decarconylase/synthase m the genome database; a gapped-BLAST (Altschul et al . (1990) J. Mol . Biol , 215:403-410) search of the non-redundant database showed that it has significant similarity co carbon monoxide dehydrogenases from other organisms, as well as several hypothetical proteins.
  • MJECS06 a hypothetical protein found by the instant FSD, exhibited similarity to other hypothetical proteins and very distant sequence similarity to zmc finger proteins.
  • MJ1552 A related protein m Methanococcus , MJECL27, exhibited conservation of several cysteines and histidmes, making it likely that this protein is also a zmc finger or related metal binding protein. None of these predicted proteins, except MJ1552, was recognized by the local sequence motif databases. MJ1552 matched one of three glutaredox patterns (PR00160A) in the Prints database using the Blocks scoring function. In summary, application of the oxidoreductase FSD to predicted models produced from all M. jannascmi ORFs found the single "true positive" m this genome (based on known biological activity) . An additional five sequences were also identified as having the sub ect active site through application of the instant FSD.
  • the assignments were ranked as “clear,” “tentative,” “marginal,” or “no assignment,” depending on the level of confidence of that data.
  • the FSD-based approach described above differs fundamentally from the GeneQuiz approach, m that three-dimensional data are used to predict active site residues.
  • the FSD approach described herein takes into account that a protein could have multiple active sites "function.”
  • the GeneQuiz analysis for the 24 sequences identified using the FSD are also shown m Table 11, above. The need for active site descriptors is clearly demonstrated by HI1159. This sequence was identified as a thioredoxin by GeneQuiz, probably because of the sequence similarity to another thioredoxin-like protein.
  • no cysteines were present m the sequence; thus, the instant FSD would not match against, or compare favorably with, this sequence as a disulfide oxidoreductase.
  • GeneQuiz identified HI1165 as a possible glutaredoxm, probably because of limited sequence identity to glutaredox -like proteins.
  • This sequence which aligned to lego by the threading algorithm, has a single cysteme, but lacks the C-X-X-C sequence; consequently it is not identified by the disulfide oxidoreductase FSD. While it has been shown that mutants m disulfide oxidoreductase protein family that lack the second cysteine can perform disulfide isomerization with limited activity, they cannot undergo oxidoreductase activity. Thus, it is unlikely that this protein can undergo the full range of reactions found m the majority of tne thiol/disulfide oxidoreductases, even though t is identified as a glutaredoxm by GeneQuiz.
  • GeneQuiz also identified MJ0530 as a member of tnis family. This protein contains two C-X-X-C sequences, but neither gapped-BLAST nor threading program algin it to any members of the glutaredoxm/thioredoxm family. BLAST only aligned it to hypothetical proteins and a possible zmc-fmger protein.
  • FSDs geometric and conformational descriptors of protein functional sites (here, disulfide oxidoreductase active sites)
  • disulfide oxidoreductase active sites are inherently more exact than local sequence signatures and provide an unambiguous mapping of function of biological FSDs to overcome the problem of using one-dimensional sequence information to recognize three-dimensional structure. They also avoid the disadvantages of sequence alignment of distantly related proteins and the problems of adequate alignment of three-dimensional structures.
  • FSD generation requires that a structure of a protein known to possess the particular biological function must be known, the resultant FSD(s) for the particular function can be applied to structures of less than high resolution.
  • FSDs according to the invention can be applied to inexact models of protein structure, including those produced by current state- of-the-art tertiary structure prediction algorithms. Furthermore, the results described m this example further support the application of FSDs m the scaled-up analysis of complete genomes.
  • M. jannaschii genome the organism's known disulfide oxidoreductase was correctly identified and five other proteins predicted to have structures of at least a portion of which satisfied the constraints of the disulfide oxidoreductase FSD.
  • FSDs can be applied to any three-dimensional structure, whether it is an atomic resolution structure or an inexact, predicted model.
  • libraries of FSDs will be useful for the massive structure- determination pro ects required m connection with genome sequencing projects, for example.
  • m one application FSDs can be used to significantly narrow down the number of ORFs that might encode a particular active site. Moreover, this procedure can be automated.
  • Protein function and structure prediction cy nomoiogy to already cnaracterized proteins is jnown Severs, programs, such as BLAST or FASTA, are geared toward recognition of protein homology by analysis of sequence similarities.
  • the first approach .e., threading
  • the second approach uses sequences of closely related proteins to estimate the patterns of mutations along the sequence and to create (a) position-specific mutation matrix.
  • the objective of this approach is the same as in conventional sequence alignment methods -- to identify homologies between families.
  • threading has a much wider application than the profile, or any other sequence-only type approach.
  • sequence- based methods can achieve more than structure prediction.
  • Mycoplasma geni talium is a small, pathogenic Gram- positive bacterium associated with pulmonary and urogenital infections m humans. Its close cousin, Mycoplasma pneumoniae, causes primary atypical pneumonia.
  • M. geni talium is a very simple organism lacking a cell wall and is the smallest known cellular organism capable of independent replication. Its entire genome, composed of 468 predicted open reading frames, was sequenced by "shotgun" sequencing and made available on the WEB site www. tigr.org, together with annotations identifying the function of about 67% of all proteins m this genome.
  • Botn PSI-BLAST and BASIC were used m two different tasks. First, structural predictions for proteins from the Mycoplasma geni talium genome were made by comparing their sequences against sequences and sequence profiles of proteins with known structures. In this application, both programs were used merely as fold prediction tools. These structure prediction results were then used to assign M. geni talium proteins to nomologous superfamilies . The same proteins were then compared against proteins from the E. coli genome to determine wnat functional assignments could be transferred between genomes . (c) Results
  • the PSI- BLAST algontrvr detected 118 significant (E value lower than 0.1) similarities to proteins with known structures. This constitutes 25% c the entire genome.
  • Fischer & Eisenberg reported 75 significant similarities smg the older version of the BLAST algorithm and a smaller version of the database of known structures [15] .
  • the BASIC program detected 176 significant (Z-score larger than 10) similarities to proteins with known structures (38%) , an almost 50% increase over the PSI-BLAST recognition rate. This was a superset of BLAST predictions, because all the high significance BLAST predictions were independently recognized by the BASIC algorithm. Thus, there BASIC provided 58 additional structural assignments. 28 of these are for hypothetical proteins, and 16 are for proteins of M. geni talium whose functions are known. Several other M. geni talium proteins, such as Arg, Phe, Cys and Trp-tRNA synthetases, undine kinase, and uracil phosphoribosyltransferase, have obvious functional similarities to the proteins identified by the BASIC algorithm.
  • the BASIC algorithm povides several new insights into the metabolism and other processes m M. geni talium. For instance:
  • a second enzyme involved m ammo acid metabolism was identified by homology with glycme methyltransferase .
  • M. geni talium is not sensitive to penicillin or other ⁇ -lactam containing ntibiotics (it lacks a cell wall) and no penicillin binding proteins have been found m Mycoplasma [28] , the problem of antibiotic resistance m multi-organism infection is not well understood. As an opportunistic pathogen, M. geni talium could degrade antibiotics as a part of symbiotic relation with other pathogens .
  • the identification of distant evolutionary relationships is a reliable structure and function prediction tool.
  • the position-specific iterative BLAST algorithm represents sensitive, conventional algorithm for such identification.
  • PSI-BLAST Materials and Methods.
  • the position specific iterative BLAST algorithm [16] is the newest version of the de facto standard of database protein similarity search algorithms. This algorithm addresses the principal shortcoming of the previous BLAST algorithm: its inability to introduce gaps m the alignment.
  • the PSI-BLAST algorithm allows the iterative building of a sequence profile from the multiple alignment of sequences of homologous protein identified the first pass of the algorithm.
  • the PSI-BLAST program was downloaded from the NIH WEB site and used following the guidelines m the manual.
  • sequence database used by the PSI-BLAST algorithm contains a non-redundant compilation of sequences available from SWISSPROT and PIR databases, as well as translated DNA sequences from EMBL and NCBI nucleotide sequence databases and sequences of all proteins deposited m the Brookhaven PDB database. The version used this work was compiled m November 1997.
  • Profile sequence preparation Profiles were generated automatically using the multiple alignment of homologous sequences as generated by the PSI-BLAST algorithm. The same procedure was followed for the target proteins as well as for all proteins contained m the databases being searched.
  • the first database of 1151 representative protein structures was prepared on the basis of a non-redundant set of protein structures included m the FSSP database as available from the DALI server at EBI . This database was used for fold prediction.
  • the second database consists of sequence profiles for all proteins from the E. coli genome, as available on the E . coli WEB site at U. Wisconsin Genome Center.
  • the BASIC algorithm was optimized to recognize the maximal number of structurally similar proteins on benchmarks customized for fold prediction algorithms.
  • a particular benchmark available from the WEB server at UCLA was used during the development of a BASIC algorithm.
  • This benchmark consists of 68 target proteins for which the correct template (structurally similar protein) has to be found m a database of ca . 300 examples.
  • Scores of individual profile-profile comparisons were corrected for size of the proteins being compared [30, 31] and used to calculate the distributions of scores for a given prediction target. The standard deviation of the distribution is calculated and used to rescale the distribution. Re-scaled scores can be easily recalculated into the prooability of the score happening by chance, which would then be equivalent to the p-value reported by BLAST.
  • the estimation of the reliability of the prediction was based on a Z-score statistic. A conservative cutoff of 10 standard deviations above the mean score was used m this work The biggest Z-score for a false positive the UCLA benchmark described above was equal to 5.2. However, bigger database sizes increase the chance of high scoring alignments of random protein pairs.

Landscapes

  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Chemical & Material Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biotechnology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Medical Informatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Theoretical Computer Science (AREA)
  • Molecular Biology (AREA)
  • Analytical Chemistry (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Proteomics, Peptides & Aminoacids (AREA)
  • Urology & Nephrology (AREA)
  • Genetics & Genomics (AREA)
  • Biomedical Technology (AREA)
  • Hematology (AREA)
  • Immunology (AREA)
  • Cell Biology (AREA)
  • Microbiology (AREA)
  • Food Science & Technology (AREA)
  • Medicinal Chemistry (AREA)
  • Biochemistry (AREA)
  • General Physics & Mathematics (AREA)
  • Pathology (AREA)
  • Peptides Or Proteins (AREA)

Abstract

L'invention porte sur des procédés et systèmes de prédiction des fonctions biologiques de protéines se basant sur le développement de descripteurs de sites fonctionnels correspondant à des fonctions biologiques distinctes des protéines. Lesdits descripteurs de sites fonctionnels sont des représentations géométriques en 3 D des sites fonctionnels des protéines pouvant également inclure des paramètres additionnels tels que des informations de conformation. Une fois développés, on utilise un ou plusieurs de ces descripteurs de sites fonctionnels (pour une ou plusieurs fonctions biologique différentes) pour tester les structures de protéines et déterminer si elles contiennent les sites fonctionnels décrits par les descripteurs correspondants. Si c'est le cas, les protéines contenant lesdits sites fonctionnels sont considérées comme ayant les fonctions biologiques correspondantes. Dans les exécutions préférées, on utilise une bibliothèque de descripteurs de sites fonctionnels pour vérifier les structures impropres de protéines, dérivant par des méthodes informatiques d'informations sur les séquences d'acides aminés, pour prédire les fonctions biologiques desdites séquences et des gènes codant pour elles.
PCT/US1999/011911 1998-05-27 1999-05-27 Procedes et systeme de prediction des fonctions biologiques de proteines WO1999061654A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
AU42185/99A AU4218599A (en) 1998-05-27 1999-05-27 Methods and systems for predicting protein function

Applications Claiming Priority (6)

Application Number Priority Date Filing Date Title
US8711098P 1998-05-27 1998-05-27
US60/087,110 1998-05-27
US9930098P 1998-08-25 1998-08-25
US60/099,300 1998-08-25
US12031199P 1999-02-16 1999-02-16
US60/120,311 1999-02-16

Publications (1)

Publication Number Publication Date
WO1999061654A1 true WO1999061654A1 (fr) 1999-12-02

Family

ID=27375617

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US1999/011911 WO1999061654A1 (fr) 1998-05-27 1999-05-27 Procedes et systeme de prediction des fonctions biologiques de proteines

Country Status (2)

Country Link
AU (1) AU4218599A (fr)
WO (1) WO1999061654A1 (fr)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2163366A1 (es) * 2000-01-27 2002-01-16 Univ Madrid Complutense Enzima electronico, con sitio activo reconfigurable.
EP1271343A1 (fr) * 2001-06-22 2003-01-02 Kelman Gesellschaft für Geninformation mbH Procédé et dispositif pour traiter et visualiser automatiquement d'ensembles d'informations
US7133864B2 (en) 2001-08-23 2006-11-07 Syngenta Participations Ag System and method for accessing biological data
CN103215228A (zh) * 2003-04-17 2013-07-24 斯克利普斯研究院 扩展真核生物遗传密码
US9797908B2 (en) 2003-06-18 2017-10-24 The Scripps Research Institute Unnatural reactive amino acid genetic code additions

Non-Patent Citations (9)

* Cited by examiner, † Cited by third party
Title
CHAO K.-M., ET AL.: "ALIGNING TWO SEQUENCES WITHIN A SPECIFIED DIAGONAL BAND.", CABIOS COMPUTER APPLICATIONS IN THE BIOSCIENCES., IRL PRESS,OXFORD., GB, vol. 08., no. 05., 1 January 1992 (1992-01-01), GB, pages 481 - 487., XP002923806, ISSN: 0266-7061 *
CLARKE N. D., ET AL.: "EVIDENCE FROM CASSETTE MUTAGENESIS FOR A STRUCTURE-FUNCTION MOTIF IN A PROTEIN OF UNKNOWN STRUCTURE.", SCIENCE, AMERICAN ASSOCIATION FOR THE ADVANCEMENT OF SCIENCE, US, vol. 240., 22 April 1988 (1988-04-22), US, pages 521 - 523., XP002923803, ISSN: 0036-8075, DOI: 10.1126/science.3282306 *
ESCALIER V., ET AL.: "PAIRWISE AND MULTIPLE IDENTIFICATION OF THREE-DIMENSIONAL COMMON SUBSTRUCTURES IN PROTEINS.", JOURNAL OF COMPUTATIONAL BIOLOGY., MARY ANN LIEBERT INC., US, vol. 05., no. 01., 1 January 1998 (1998-01-01), US, pages 41 - 56., XP002923802, ISSN: 1066-5277 *
KALHAMMER G., ET AL.: "RAS-BINDING DOMAINS: PREDICTING FUNCTION VERSUS FOLDING.", FEBS LETTERS., ELSEVIER, AMSTERDAM., NL, vol. 414., 1 January 1997 (1997-01-01), NL, pages 599 - 602., XP002923805, ISSN: 0014-5793, DOI: 10.1016/S0014-5793(97)01076-4 *
PAWLOWSKI K., ET AL.: "THE HELICOBACTER PYLORI GENOME: FROM SEQUENCE ANALYSIS TO STRUCTURAL AND FUNCTIONAL PREDICTIONS.", PROTEINS: STRUCTURE, FUNCTION, AND BIOINFORMATICS, JOHN WILEY & SONS, INC., US, vol. 36., 1 July 1999 (1999-07-01), US, pages 20 - 30., XP002923808, ISSN: 0887-3585, DOI: 10.1002/(SICI)1097-0134(19990701)36:1<20::AID-PROT2>3.0.CO;2-X *
RYCHLEWSKI L., ET AL.: "FUNCTIONAL INSIGHTS FROM STRUCTURAL PREDICTIONS: ANALYSIS OF THE ESCHERICHIA COLI GENOME.", PROTEIN SCIENCE, WILEY, US, vol. 08., 1 March 1999 (1999-03-01), US, pages 614 - 624., XP002923807, ISSN: 0961-8368 *
VISWANADHAN V. N., ET AL.: "NEW JOINT PREDICTION ALGORITHM (Q1-JASEP) IMPROVES THE PREDICTION OF PROTEIN SECONDARY STRUCTURE.", BIOCHEMISTRY, AMERICAN CHEMICAL SOCIETY, US, vol. 30., no. 46., 1 January 1991 (1991-01-01), US, pages 11164 - 11172., XP002923804, ISSN: 0006-2960, DOI: 10.1021/bi00110a021 *
WALLACE A. C., LASKOWSKY R. A., THORNTON J. M.: "DERIVATION OF 3D COORDINATE TEMPLATES FOR SEARCHING STRUCTURAL DATABASES: APPLICATION TO SER-HIS-ASP CATALYTIC TRIADS IN THE SERINE PROTEINASES AND LIPASES.", PROTEIN SCIENCE, WILEY, US, 1 January 1996 (1996-01-01), US, pages 1001 - 1013., XP002923801, ISSN: 0961-8368 *
ZHANG B., ET AL.: "FROM FOLD PREDICTIONS TO FUNCTION PREDICTIONS: AUTOMATION OF FUNCTIONAL SITE CONSERVATION ANALYSIS FOR FUNCTIONAL GENOME PREDICTIONS.", PROTEIN SCIENCE, WILEY, US, vol. 08., 1 March 1999 (1999-03-01), US, pages 1104 - 1115., XP002923809, ISSN: 0961-8368 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
ES2163366A1 (es) * 2000-01-27 2002-01-16 Univ Madrid Complutense Enzima electronico, con sitio activo reconfigurable.
EP1271343A1 (fr) * 2001-06-22 2003-01-02 Kelman Gesellschaft für Geninformation mbH Procédé et dispositif pour traiter et visualiser automatiquement d'ensembles d'informations
US7133864B2 (en) 2001-08-23 2006-11-07 Syngenta Participations Ag System and method for accessing biological data
CN103215228A (zh) * 2003-04-17 2013-07-24 斯克利普斯研究院 扩展真核生物遗传密码
US9580721B2 (en) 2003-04-17 2017-02-28 The Scripps Reserach Institute Expanding the eukaryotic genetic code
US9797908B2 (en) 2003-06-18 2017-10-24 The Scripps Research Institute Unnatural reactive amino acid genetic code additions

Also Published As

Publication number Publication date
AU4218599A (en) 1999-12-13

Similar Documents

Publication Publication Date Title
US6631332B2 (en) Methods for using functional site descriptors and predicting protein function
Schauperl et al. AI-based protein structure prediction in drug discovery: impacts and challenges
Fetrow et al. Method for prediction of protein function from sequence using the sequence-to-structure-to-function paradigm with application to Glutaredoxins/Thioredoxins and T1Ribonucleases
Keskin et al. Hot regions in protein–protein interactions: the organization and contribution of structurally conserved hot spot residues
US7139665B2 (en) Computational method for designing enzymes for incorporation of non natural amino acids into proteins
Fiser et al. Modeling of loops in protein structures
Fiser Protein structure modeling in the proteomics era
Ming et al. A review of enzyme design in catalytic stability by artificial intelligence
US20030130797A1 (en) Protein modeling tools
Ring et al. Modeling protein structures: construction and their applications
Gromiha Protein interactions: Computational methods, analysis and applications
WO1999061654A1 (fr) Procedes et systeme de prediction des fonctions biologiques de proteines
Fetrow et al. The protein folding problem: a biophysical enigma
Gromiha Distinct roles of conventional non-covalent and cation–π interactions in protein stability
Schenk et al. Protein sequence and structure alignments within one framework
Bespamyatnikh et al. Accurate protein docking by shape complementarity alone
US20030049687A1 (en) Novel methods for generalized comparative modeling
Olson et al. Enhancing sampling of the conformational space near the protein native state
Fiser Comparative protein structure modelling
Srikanth et al. Advances in the computational methods to understand protein folding and stability
Opuu Computational design of proteins and enzymes
Zhang et al. Overview of structural bioinformatics
Krebs et al. Studying protein flexibility in a statistical framework: tools and databases for analyzing structures and approaches for mapping this onto sequences
Jia Using evolutionary covariance to infer protein sequence-structure relationships
Kaufmann Computational prediction of protein small molecule interfaces using ROSETTA

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): AE AL AM AT AU AZ BA BB BG BR BY CA CH CN CU CZ DE DK EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MD MG MK MN MW MX NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT UA UG US US US UZ VN YU ZA ZW

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): GH GM KE LS MW SD SL SZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
NENP Non-entry into the national phase

Ref country code: KR

REG Reference to national code

Ref country code: DE

Ref legal event code: 8642

122 Ep: pct application non-entry in european phase