A prediction method for DNA binding residues based on a three-segment sliding window

Abstract

A method for predicting DNA binding residues based on a three-segment sliding window is disclosed. Firstly, according to the input protein sequence information P for DNA binding residue prediction, the number of residues is L, the protein sequence database UniRef90 is searched using the PSI-BLAST program to generate a position-specific scoring matrix of size L×20, a solvent accessibility feature matrix of size L×3 is obtained using the SANN program, and a feature matrix of size L×20 is obtained by one-hot encoding. Then, the three feature matrices are concatenated into a super feature matrix of size L×43, that is, each protein residue corresponds to a 1×43 feature vector, and the similarity of the feature vectors between two residues is obtained by cosine similarity, that is, the residue similarity. Secondly, the feature view processing is performed on each residue in the super matrix by using the three-segment sliding window technology, and these feature views are expanded into feature vectors. Thirdly, the corresponding feature vectors and labels are generated by using the protein sequence of known DNA binding residues, and input into the support vector machine for training. Finally, the feature vector generated by the protein sequence to be predicted is input into the trained model to predict whether the residues in the protein sequence are DNA binding residues. The present invention has low calculation cost and high prediction accuracy.

Description

Technical Field

[0001] The invention relates to the fields of bioinformatics, pattern recognition and computer application, and in particular to a method for predicting DNA binding residues based on a three-segment sliding window. Background Art

[0002] Prediction of protein-ligand binding is an important method to promote drug development and biotechnology applications in the post-gene era. DNA molecules are an important type of ligand molecules in biology. Accurately identifying the residues in protein sequences that bind to DNA molecules is conducive to understanding protein function, analyzing the interaction mechanism between proteins and DNA molecules, and drug development. Using sliding windows is a common method in the feature processing stage of protein-ligand binding prediction.

[0003] The literature review found that many methods for predicting DNA binding residues in protein sequences use sliding windows in the feature processing stage, such as: Zhou J, Lu Q, Xu R, et al. CNNsite: prediction of DNA-binding residues in proteins using convolutional neural network with sequence features[J], 2016: 78-85. That is: Zhou J et al. Prediction of protein DNA binding residues based on convolutional neural network based on sequence features[J], 2016: 78-85, Zhang Y, Qiao S, Ji S, et al. Identification of DNA-protein binding sites by bootstrap multiple convolutional neural networks by onsequence information[J]. Engineering Applications of Artificial Intelligence, 2019; 79(March): 58–66. That is: Zhang Y et al. Identification of DNA-protein binding sites by bootstrap multiple convolutional neural networks by onsequence information[J]. Engineering Applications of Artificial Intelligence, 2019; 79(3): 58-66, etc. Although existing methods use sliding windows to predict DNA binding residues in protein sequences, the prediction accuracy cannot be guaranteed to be optimal by relying on the traditional sliding window approach and needs to be further improved.

[0004] In summary, the existing prediction methods for DNA binding residues are still far from practical application in terms of computational cost and prediction accuracy, and are in urgent need of improvement. Summary of the invention

[0005] In order to overcome the shortcomings of the traditional sliding window DNA binding residue prediction method in terms of computational cost and prediction accuracy, the present invention proposes a DNA binding residue prediction method with a three-segment sliding window processing feature with low computational cost and high prediction accuracy.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] A method for predicting DNA binding residues based on a three-segment sliding window, the prediction method comprising the following steps:

[0008] 1) Input a protein sequence information P with a residue number L for which DNA binding residues are to be predicted;

[0009] 2) For protein sequence P, the PSI-BLAST program is used to search the protein sequence database UniRef90 to generate a position-specific score matrix of size L×20, denoted as PSSM, and the SANN program is used to obtain a solvent accessibility feature matrix of size L×3, denoted as PSA. A feature matrix of size L×20 is obtained through one-hot encoding, denoted as PHOT;

[0010] 3) Substitute 21 obtained in step 2) to obtain a feature matrix of size L×43, denoted as F;

[0011] 4) Residue P in protein sequence P λ , λ=1,2,…,the eigenvector of L is represented by the λth row in F, denoted by F λ ;

[0012] 5) Taking the target residue P i , i=1,2,…,L as an example, the characteristic vector F of the target residue is calculated for the L×43 characteristic matrix obtained in 3). i The feature vector F of other residues j , j = 1, 2, ..., the cosine value of L represents the relationship between other residues and P i The similarity of

[0013]

[0014] P j are the residues other than the target residue on sequence P, j≠i;

[0015] 6) Find a residue P with the highest similarity to the target residue on both sides x , P y , use three sliding windows of size ∈ to extract P i , P x , P y The corresponding feature matrix, that is, the three ε×43 feature matrices, is denoted as M i 、M x 、M y ;

[0016] 7) If P i At both ends of the sequence P, a sliding window of size ∈ is used to extract the ε×43 zero feature matrix as M x or M y ;

[0017] 8) The feature matrix extracted from the three windows is expressed as M i As the center, spliced ​​into a 3ε×43 feature matrix as P i The feature matrix of the target residue is extracted by a three-segment sliding window, and finally L 3ε×43 feature matrices are obtained. These L 3ε×43 feature matrices are processed into L 1×3ε×43 feature vectors;

[0018] 9) Obtain protein sequence information of known DNA binding residue states from the protein structure database PDB as a training protein set, generate a feature vector for each residue using steps 2)-8), and construct a training sample set;

[0019] 10) The feature vector and label are input into the support vector machine for training. Finally, the feature tensor generated by each residue of the protein sequence P is input into the trained model to output whether the corresponding residue is a DNA binding residue.

[0020] The technical concept of the present invention is as follows: first, according to the protein sequence information P to be input for DNA binding residue prediction, the number of residues is L, the PSI-BLAST program is used to search the protein sequence database UniRef90 to generate a position-specific scoring matrix of size L×20, the SANN program is used to obtain a solvent accessibility feature matrix of size L×3, and a feature matrix of size L×20 is obtained by one-hot encoding; then, the three feature matrices are spliced ​​into a super feature matrix of size L×43, that is, each residue of the protein corresponds to a 1×43 feature vector, and the similarity of the feature vectors between two residues is obtained by cosine similarity, that is, the residue similarity; secondly, a three-segment sliding window technology is used to process the feature view of each residue in the super matrix, and these feature views are expanded into feature vectors; thirdly, the protein sequence of known DNA binding residues is used to generate the corresponding feature vector and label input support vector machine for training; finally, the feature vector generated by the protein sequence to be predicted is input into the trained model to predict whether the residue in the protein sequence is a DNA binding residue.

[0021] The beneficial effects of the present invention are as follows: on the one hand, the three-segment sliding window aims to not only obtain the characteristic information of the residues adjacent to the target residue that have a strong ability to act on it from the characteristics of the residues, but also to find residues with high similarity to the characteristics of the target residues in the entire protein to improve the prediction accuracy; on the other hand, a simple support vector machine is used to predict DNA binding residues, which further improves the prediction efficiency and accuracy of DNA binding residues. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Schematic diagram of a DNA binding residue prediction method based on a three-segment sliding window.

[0023] Figure 2 This is the result of predicting the binding residue information of protein 3J9X_F using a three-segment sliding window DNA binding residue prediction method. DETAILED DESCRIPTION

[0024] The present invention will be further described below in conjunction with the accompanying drawings.

[0025] Reference Figure 1 and Figure 2 , a DNA binding residue prediction method based on a three-segment sliding window, comprising the following steps:

[0026] 1) Input a protein sequence information P with a residue number L for which DNA binding residues are to be predicted;

[0027] 2) For protein sequence P, the PSI-BLAST program is used to search the protein sequence database UniRef90 to generate a position-specific score matrix of size L×20, denoted as PSSM, and the SANN program is used to obtain a solvent accessibility feature matrix of size L×3, denoted as PSA. A feature matrix of size L×20 is obtained through one-hot encoding, denoted as PHOT;

[0028] 3) Concatenate the PSSM, PHOT and PSA features obtained in step 2) in the first dimension to obtain a feature matrix of size L×43, denoted as F;

[0029] 4) Residue P in protein sequence P λ , λ=1,2,…,the eigenvector of L is represented by the λth row in F, denoted by F λ ; 5) with the target residue P i , i=1,2,…,L as an example, the characteristic vector F of the target residue is calculated for the L×43 characteristic matrix obtained in 3). i The feature vector F of other residues j , j = 1, 2, ..., the cosine value of L represents the relationship between other residues and P i The similarity of

[0030]

[0031] P j are the residues other than the target residue on sequence P, j≠i;

[0032] 6) Find a residue P with the highest similarity to the target residue on both sides x , P y , use three sliding windows of size ∈ to extract P i , P x , P y The corresponding feature matrix, that is, the three ε×43 feature matrices, is denoted as M i 、M x 、M y ;

[0033] 7) If P i At both ends of the sequence P, a sliding window of size ∈ is used to extract the ε×43 zero feature matrix as M x or M y ;

[0034] 8) The feature matrix extracted from the three windows is expressed as M i As the center, spliced ​​into a 3ε×43 feature matrix as P iThe feature matrix of the target residue is extracted by a three-segment sliding window, and finally L 3ε×43 feature matrices are obtained. These L 3ε×43 feature matrices are processed into L 1×3ε×43 feature vectors;

[0035] 9) Obtain protein sequence information of known DNA binding residue states from the protein structure database PDB as a training protein set, use steps 2)-8) to generate a feature vector for each residue, and construct a training sample set;

[0036] 10) The feature vector and label are input into the support vector machine for training. Finally, the feature tensor generated by each residue of the protein sequence P is input into the trained model to output whether the corresponding residue is a DNA binding residue.

[0037] This example takes the prediction of DNA binding residues of protein 3J9X_F as an example, and a method for predicting DNA binding residues based on a three-segment sliding window includes the following steps:

[0038] 1) Input a protein sequence 3J9X_F with 128 residues to be predicted for DNA binding residues;

[0039] 2) For protein sequence 3J9X_F, the PSI-BLAST program was used to search the protein sequence database UniRef90 to generate a 128×20 position-specific score matrix, denoted as PSSM, and the SANN program was used to obtain a 128×3 solvent accessibility feature matrix, denoted as PSA. A 128×20 feature matrix was obtained through one-hot encoding, denoted as PHOT;

[0040] 3) Concatenate the PSSM, PHOT and PSA features obtained in step 2) in the first dimension to obtain a feature matrix of size 128×43, denoted as F;

[0041] 4) Residue P in protein sequence 3J9X_F λ , the eigenvector of λ=1,2,…,128 is represented by the λth row in F, denoted as F λ ;

[0042] 5) Taking the target residue P i , i=1,2,…,128 as an example, the feature vector F of the target residue is calculated for the 128×43 feature matrix obtained in 3). i The feature vector F of other residues j , j = 1, 2, ..., 128 to represent the cosine value of other residues and P i The similarity of

[0043]

[0044] P j are the residues other than the target residue on sequence P, j≠i;

[0045] 6) Find a residue P with the highest similarity to the target residue on both sides x , P y , use three sliding windows of size 7 to extract P i , P x , P y The corresponding feature matrix, that is, three 7×43 feature matrices, is denoted as M i 、M x 、M y ;

[0046] 7) If P i At both ends of the sequence 3J9X_F, a sliding window of size ∈ is used to extract a 7×43 zero feature matrix as M x or M y ;

[0047] 8) The feature matrix extracted from the three windows is expressed as M i As the center, spliced ​​into a 21×43 feature matrix as P i The feature matrix of the target residues is extracted by a three-segment sliding window, and finally 128 feature matrices of 21×43 are obtained. These 128 feature matrices of 21×43 are processed into 128 feature vectors of 1×21×43.

[0048] 9) Obtain protein sequence information of known DNA binding residue states from the protein structure database PDB as a training protein set, use steps 2)-8) to generate a feature vector for each residue, and construct a training sample set;

[0049] 10) The feature vector and label are input into the support vector machine for training. Finally, the feature vector generated by each residue in the protein sequence 3J9X_F is input into the trained model, and the output is whether the corresponding residue is a DNA binding residue.

[0050] In this example, the feature vector generated by each residue in the protein sequence 3J9X_F is input into the trained model, and the number of DNA binding residues in the protein sequence 3J9X_F is obtained to be 15.

Claims

1. A DNA binding residue prediction method based on a three-segment sliding window, characterized in that: The prediction method comprises the following steps: 1) Input a protein sequence information P with a residue number L for which DNA binding residues are to be predicted; 2) For protein sequence P, the PSI-BLAST program is used to search the protein sequence database UniRef90 to generate a position-specific score matrix of size L×20, denoted as PSSM, and the SANN program is used to obtain a solvent accessibility feature matrix of size L×3, denoted as PSA. A feature matrix of size L×20 is obtained through one-hot encoding, denoted as PHOT; 3) Concatenate the PSSM, PHOT and PSA features obtained in step 2) in the first dimension to obtain a feature matrix of size L×43, denoted as F; 4) Residue P in protein sequence P λ , λ=1,2,…,the eigenvector of L is represented by the λth row in F, denoted by F λ ; 5) Taking the target residue P i , i=1,2,…,L as an example, the characteristic vector F of the target residue is calculated for the L×43 characteristic matrix obtained in 3). i The feature vector F of other residues j , j = 1, 2, ..., the cosine value of L represents the relationship between other residues and P i The similarity of P j are the residues other than the target residue on sequence P, j≠i; 6) Find a residue P with the highest similarity to the target residue on both sides x , P y , use three sliding windows of size ∈ to extract P i , P x , P y The corresponding feature matrix, that is, the three ε×43 feature matrices, is denoted as M i 、M x 、M y ; 7) If P i At both ends of the sequence P, a sliding window of size ∈ is used to extract the ε×43 zero feature matrix as M x or M y ; 8) The feature matrix extracted from the three windows is expressed as M i As the center, spliced ​​into a 3ε×43 feature matrix as P i The feature matrix of the target residue is extracted by a three-segment sliding window, and finally L 3ε×43 feature matrices are obtained. These L 3ε×43 feature matrices are processed into L 1×3ε×43 feature vectors; 9) Obtain protein sequence information of known DNA binding residue states from the protein structure database PDB as a training protein set, use steps 2)-8) to generate a feature vector for each residue, and construct a training sample set; 10) The feature vector and label are input into the support vector machine for training. Finally, the feature tensor generated by each residue of the protein sequence P is input into the trained model to output whether the corresponding residue is a DNA binding residue.

Images