CN113764045B - DNA-binding protein identification method and related products based on XGboost algorithm - Google Patents

Abstract

A DNA-binding protein identification method, system, storage medium and device based on the XGboost algorithm belong to the technical field of computer and protein identification and combination. The present invention solves the problem that the existing DNA-binding protein identification methods cannot take into account both versatility and identification accuracy. The invention uses the DNA binding protein identification classifier to identify the DNA binding protein to be identified; in the determination process of the DNA binding protein identification classifier, firstly obtains the processed DNA binding protein characteristic data set; uses different extraction algorithms to extract the DNA binding protein data Then, the generated feature matrix is normalized, and the MRMD algorithm matrix is used for dimensionality reduction processing. ; Finally, the XGboost algorithm was used to construct and train the DNA-binding protein recognition classifier model. Mainly used for the identification of DNA-binding proteins.

Description

DNA-binding protein identification method and related products based on XGboost algorithm

technical field

The invention belongs to the technical field of computer and protein identification and combination, and in particular relates to a DNA-binding protein identification method, system, storage medium and device.

Background technique

Organisms contain many macromolecular substances, such as DNA and proteins, which contain the genetic information of the organism and are an important part of all cells and tissues that make up the organism. In order to study the life activities of cells, it is necessary to study DNA and proteins and the interactions between them. The study of DNA-binding proteins plays an important role in DNA replication and recombination, virus infection and proliferation. To study the gene expression of an organism at the molecular level, it is necessary to study the binding of DNA and proteins. Therefore, the accurate identification of DNA-binding proteins is the premise of further research on the life activities of cells.

The current common detection methods are usually single or complex, that is, a feature extraction method and a training model or a more complex algorithm such as a convolutional neural network is used to identify DNA-binding proteins. Different feature extraction methods have different emphases, and the recognition results are also different. Therefore, the existing DNA-binding protein recognition methods cannot take into account the generality and recognition accuracy.

SUMMARY OF THE INVENTION

In order to solve the problem that the existing DNA binding protein identification methods cannot take into account the versatility and the identification accuracy, the present invention further proposes a DNA binding protein identification method based on the XGboost algorithm, so as to realize more accurate identification of DNA binding proteins at the same time. Improve versatility.

The DNA-binding protein identification method based on the XGboost algorithm uses the DNA-binding protein identification classifier to identify the DNA-binding protein to be identified, and the determination process of the DNA-binding protein identification classifier includes the following steps:

S1. Obtain the processed DNA-binding protein feature data set; the DNA-binding protein feature data set includes a training set and a test set;

S2, using different extraction algorithms to extract the data features of the DNA-binding protein dataset to obtain multiple feature files;

S3, splicing the sequence feature matrices extracted by different feature extraction algorithms to obtain a spliced feature matrix;

S4, normalize the feature matrix generated by S3 to obtain a normalized feature matrix;

S5. Use the MRMD algorithm to perform dimension reduction processing on the matrix generated by S4;

S6. Use the XGboost algorithm to build and train a DNA-binding protein recognition classifier model.

Further, S2 uses different extraction algorithms to extract the data features of the original DNA-binding protein dataset. The extraction algorithms used are global encoding method of protein sequence, Multi-scaleContinuous and Discontinuous, Novel Matrix-Based Sequence Representation Model with Amino Acid, Position-Specific Scoring Matrix PSSM-AB, PSSM-Pse and PSSM-DWT.

Further, the normalization process described in S4 uses a zero-mean normalization algorithm.

Further, the MRMD algorithm adopts the MRMD3.0 algorithm.

A DNA-binding protein identification system based on the XGboost algorithm, the system is used for executing the DNA-binding protein identification method based on the XGboost algorithm.

A storage medium, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to realize the DNA-binding protein identification method based on the XGboost algorithm.

A device, the device includes a processor and a memory, the memory stores at least one instruction, the at least one instruction is loaded and executed by the processor to implement the DNA-binding protein identification method based on the XGboost algorithm.

The beneficial effects of the present invention are:

The invention utilizes the composition of ribonucleotides to express the characteristics of the protein sequence, can realize the accurate identification of the DNA binding protein, and provides a theoretical basis for the corresponding research and development. Moreover, when the present invention constructs the model, the feature matrix is spliced and standardized, which effectively strengthens the connection between the data feature and the data label, thereby improving the accuracy. On this basis, the present invention uses the MRMD3.0 algorithm to reduce the dimension of the data, obtains higher recognition accuracy with less data, and improves the recognition efficiency of DNA-binding proteins.

The present invention builds a classifier through the XGboost algorithm, and generates a DNA binding protein identification model with better comprehensive performance and wider application, and the present invention uses different feature extraction methods to extract features and identify them based on different features, so the DNA binding can be further improved. The versatility and recognition accuracy of the protein recognition method, that is, different proteins can be recognized and guaranteed to have a very high accuracy. Further, the present invention innovatively uses the feature sets extracted by 6 kinds of feature extraction methods and reduces the dimension, combines the feature information extracted by the 6 kinds of methods, and selects the information with higher feature value through the dimensionality reduction algorithm, The XGboost algorithm, which is simpler than the neural network, is used for identification. On the basis of generality, it can achieve a higher identification effect with fewer feature sets and simpler algorithms.

Description of drawings

Fig. 1 is the flow chart of the DNA-binding protein identification method based on machine learning;

Figure 2 shows the comparison results on the unstitched independent dataset and the stitched independent dataset.

Detailed ways

Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the accompanying drawings are exemplary only, and are intended to illustrate the principles and spirit of the present invention, and not to limit the scope of the present invention.

Specific implementation one:

The present embodiment is a DNA-binding protein identification method based on the XGboost algorithm, as shown in Figure 1, including the following steps:

S1. Download a DNA-binding protein sequence data file to obtain a processed DNA-binding protein feature data set.

In this embodiment, two training sets and two test sets are used to train and test the model. Among them, the training set PDB1075 and the test set PDB186 and the training set PDB14189 and the test set PDB2272 are downloaded from the ProteinDatabase database.

The training dataset PDB1075 consists of 525 DNA-binding proteins and 550 non-DNA-binding proteins, and the test dataset PDB186 consists of 93 DNA-binding proteins and 93 non-DNA-binding proteins. Each protein consists of 100 amino acids, and the training dataset PDB14189 Consisting of 7129 DNA-binding proteins and 7060 non-DNA-binding proteins, the test dataset PDB2272 consists of 1153 DNA-binding proteins and 1119 non-DNA-binding proteins with protein sequence similarity less than 30%.

S2. Use different extraction algorithms to extract the data features of the original DNA-binding protein data set to obtain multiple feature files; in the process of extracting the data features of the original DNA-binding protein data set by using different extraction algorithms, different features The extraction algorithm extracts the same sequence.

In this example, six feature extraction methods are used, including global encoding method of protein sequence (GE), Multi-scale Continuous and Discontinuous (MCD), NovelMatrix-Based Sequence Representation Model with Amino Acid (NMBAC), Position-Specific Scoring Matrix (PSSM) PSSM-AB, PSSM-Pse and PSSM-DWT.

The statistical principle of GE is obtained by global coding; MCD obtains sequence features through multi-scale continuous and discontinuous descriptors; NMBAC statistical principle is normalized Moreau-Broto autocorrelation; PSSM-AB is based on position-specific scoring matrix Average block; PSSM-DCT is PSSM-based discrete cosine transform; PSSM-DWT is PSSM-based discrete wavelet transform. Protein sequences can be represented by these discrete values.

Other numbers of feature extraction methods and other feature extraction methods may also be used in other embodiments. It should be noted that the present invention innovatively uses the feature sets extracted by these six feature extraction methods and performs dimensionality reduction in the subsequent process, and conducts in-depth research and experiments on the structure and physiological properties of different DNA-binding proteins. , it is found that the feature information extracted by these six methods can effectively represent and represent different DNA-binding proteins universally, so as to obtain a very good prediction effect (universality and recognition accuracy); that is to say, these six methods of the present invention The feature extraction method and the corresponding feature combination were discovered through in-depth research and experimental discovery on the structure and physiological properties of different DNA-binding proteins and used innovatively. The structure and physiological form must be very similar, so they will be classified as a type of protein, that is to say, the identification of the same type of protein is carried out by using its same or similar structure, so most of them are based on a certain type of protein. For feature identification, even if there is a way of using different features for protein identification, due to the influence of the existing way of thinking of the same/similar structure and properties of proteins, it is generally not possible to determine whether the features extracted by different feature extraction methods are acceptable or not. We will conduct research on better representation of proteins in different spatial dimensions, and will not study which combination of features is more conducive to the versatility and accuracy of recognition methods, and will not predict which combination of features. The method will achieve better results, that is to say: for those skilled in the art, the expected effect is only that each recognition effect of different features should be similar to the common recognition effect, that is, assuming that there are n kinds of features, respectively The corresponding effect is assumed to be 1, then the effect of common identification of these n features is at most n, and will not exceed n (because there may be correlations between different features themselves, and the features are related in the data space. ), and those skilled in the art would not expect that "different features represent different aspects, and for the overall identification of proteins, the combination of different features is more conducive to the multi-directional expression of proteins in the data space, so different features It is equivalent to having a synergistic effect between them”, and the 6 features selected after research and experiments in the present invention have very good synergistic expression, and then obtain a recognition effect exceeding n.

At the same time, these six feature extraction methods will also synergize and promote the subsequent processing process, that is, the information with higher feature value is screened out through the dimensionality reduction algorithm, and the XGboost algorithm, which is simpler than the neural network, can be used for identification. Feature sets and simpler algorithms achieve high recognition results.

S3. The feature matrix generated according to the feature file of S2: extract features for the same sequence, and splicing the sequence feature matrices extracted by different feature extraction algorithms to obtain a spliced feature matrix.

The feature matrix in this embodiment is extracted from the same sequence by six feature extraction methods, and the six sequence feature matrices extracted by these six feature extraction methods are spliced together.

Different feature extraction methods use different algorithms to extract sequence features, and each algorithm has its own pertinence and characteristics; splicing the sequence feature matrices extracted by different methods can effectively make up for the shortcomings of each method and improve the accuracy. Therefore, this algorithm is selected for data construction.

S4, using the zero-mean normalization algorithm to normalize the feature matrix generated by S3 to obtain a normalized feature matrix.

Using the zero-mean normalization algorithm to normalize the feature matrix generated by S3 can strengthen the relationship between data features and data labels, make the data more standardized and unified, and effectively improve the accuracy.

S5. Use the MRMD3.0 algorithm to perform dimension reduction processing on the matrix generated by S4.

The Max-Relevance-Max-Distance (MRMD3.0) algorithm is a maximum correlation and maximum distance dimension reduction method. A dimensionality reduction method developed by Zou Quan et al. in 2015, named Max-Relevance-Max-Distance (MRMD), the user guide and the complete runtime program can be obtained and downloaded from: https://github.com /heshida01 /MRMD3.0. Dimensionality reduction is to reduce the dimension, calculate the weight of each sequence feature through an algorithm and perform classification comparison, filter the features with low weight and discard the features with high weight, and record the results. It judges the data independence through the distance function, and completes the dimensionality reduction operation in three steps. It first evaluates the contribution of each feature to classification, and then quantifies the contribution of each feature to classification. Second, the weights of different features are calculated for classification, and the selected features are sorted accordingly. Finally, filter and classify different numbers of features, and record the results.

S6. Use the XGboost algorithm to build and train a DNA-binding protein recognition classifier model.

The XGboost algorithm is a machine learning model that achieves stronger learning effects by integrating multiple weak learners. The XGBoost model performs a second-order Taylor expansion of the loss function and uses various methods to prevent overfitting as much as possible. The feature matrix generated in step S5 is stored in a CSV file, and the XGboost algorithm is used to read the file and perform classification and recognition calculation to generate a DNA-binding protein recognition classifier.

The DNA-binding protein to be identified is identified by using the trained DNA-binding protein identification classifier.

Performance comparison of the present invention with other advanced DNA-binding protein models:

On the PDB 1075 dataset, the performance of spliced-sequence features and single-sequence features is evaluated by randomly extracting 30% of the data as the test set.

Table 1 is the dimension of the feature

Table 2 is the dataset information

Figure 2 and Tables 3-5 describe the experimental results. Compared with other single-sequence features, PSSM-DWT (mcc: 0.4981) achieves better performance. Concatenated sequence features perform better than single sequence features on all parameters. The splice sequence feature (ROC: 0.85) also achieved the best ROC performance.

Table 3 shows the processing results of the XGboost algorithm based on different feature extraction methods for the PDB1075 dataset

Table 4 shows the processing results of different methods for the PDB2272 dataset

Table 5 shows the processing results of different methods for the PDB186 test set (PDB1075 training)

We use PDB1075 as the training set and PDB186 as the test set to evaluate our experimental method, and compare the experimental results of our method with those of 13 other methods. Figure 2 clearly shows the complete experimental results. For the works of MSDBP, MSFBinder, Local-DPP MKSVM-HKA, and Adilina, the MCC values of the five methods are all above 0.6 (0.606, 0.616, 0.625, 0.648, and 0.670, respectively). Therefore, these methods have good performance. While Adilina's work (SN: 95.0%) performed the best in terms of SN value, the results of XGBoost achieved the best Acc (85.48%), MCC (0.713) and Spec (80.6%). On PDB1075 and PDB186, XGBoost outperforms other methods.

We removed proteins in PDB2272 with more than 40% sequence homology to PDB14189 to avoid homology bias between the two datasets. PDB14189 is the training set and PDB2272 is the test set. We independently tested XGBoost on PDB2272, using PDB14189 as the training set, and compared it with 5 other classification methods. The detailed experimental results are shown in Figure 2. The experimental results show that compared with other methods, XGBoost obtains the optimal ACC value, MCC value and SPEC value, which are 78.26%, 0.5652 and 76.05%, respectively. For PDB2272, XGBoost shows better performance than other classification methods.

Specific implementation two:

This embodiment is a DNA-binding protein identification system based on the XGboost algorithm, and the system is used to execute the DNA-binding protein identification method based on the XGboost algorithm.

Specific implementation three:

This embodiment is a storage medium, and the storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement the DNA-binding protein identification method based on the XGboost algorithm.

Specific implementation four:

This embodiment is a device, the device includes a processor and a memory, the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the DNA-binding protein based on the XGboost algorithm recognition methods.

The above calculation examples of the present invention are only to illustrate the calculation model and calculation process of the present invention in detail, but are not intended to limit the embodiments of the present invention. For those of ordinary skill in the art, on the basis of the above description, other different forms of changes or changes can also be made, and it is impossible to list all the embodiments here. Obvious changes or modifications are still within the scope of the present invention.

Claims (6)

1. the DNA binding protein identification method based on XGboost algorithm is characterized in that, utilizes DNA binding protein identification classifier to identify the DNA binding protein to be identified, and the determination process of described DNA binding protein identification classifier may further comprise the steps: S1. Obtain the processed DNA-binding protein feature data set; the DNA-binding protein feature data set includes a training set and a test set; S2. Use different extraction algorithms to extract the data features of the DNA-binding protein data set to obtain multiple feature files; the extraction algorithm used in the process of using different extraction algorithms to extract the data features of the original DNA-binding protein data set is the global encoding method of protein sequence, Multi-scale Continuous and Discontinuous, Novel Matrix-Based Sequence Representation Model with AminoAcid, Position-Specific Scoring Matrix PSSM-AB, PSSM-Pse and PSSM-DWT; global encoding method of protein sequence is the global encoding method of protein sequence; Multi-scale Continuous and Discontinuous is the multi-scale continuous and discontinuous method; NovelMatrix-Based Sequence Representation Model with Amino Acid is the normalized matrix amino acid sequence representation model method; Position specific scoring matrix-based average blocks is an average block method based on a specific position scoring matrix; PSSM-Pse is a local pseudo-specific position scoring matrix method; PSSM-DWT is a discrete wavelet transform method based on a specific position scoring matrix; S3, splicing the sequence feature matrices extracted by different feature extraction algorithms to obtain a spliced feature matrix; S4, normalize the feature matrix generated by S3 to obtain a normalized feature matrix; S5. Use the MRMD algorithm to reduce the dimension of the matrix generated by S4; the MRMD algorithm is the maximum correlation and maximum distance dimension reduction method; S6. Use the XGboost algorithm to build and train a DNA-binding protein recognition classifier model. 2 . The DNA-binding protein identification method based on the XGboost algorithm according to claim 1 , wherein the normalization process described in S4 uses a zero-mean normalization algorithm. 3 . 3. The DNA-binding protein identification method based on the XGboost algorithm according to claim 2, wherein the MRMD algorithm adopts the MRMD3.0 algorithm. 4. A DNA-binding protein identification system based on the XGboost algorithm, characterized in that the system is used to execute the DNA-binding protein identification method based on the XGboost algorithm described in any one of claims 1 to 3. 5. A storage medium, characterized in that, at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the XGboost algorithm-based algorithm according to one of claims 1 to 3. DNA-binding protein identification methods. 6. A DNA-binding protein identification device based on the XGboost algorithm, wherein the device comprises a processor and a memory, and at least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to achieve The DNA-binding protein identification method based on the XGboost algorithm according to any one of claims 1 to 3.

Images