Granular support vector machine ensemble-based protein ligand binding site prediction method

Abstract

The invention discloses a granular support vector machine ensemble-based protein ligand binding site prediction method. The method comprises the steps of 1, performing feature extraction according to evolution information and a secondary structure of a protein sequence, representing amino acid residues in the sequence in an eigenvector form, and constructing a training sample set by the residues (sites) as units; 2, performing sampling on the training sample set by utilizing a granular computation thought to generate a plurality of sub-training sample sets; 3, training support vector machine (SVM) models on the sub-training sample sets respectively, wherein multiple SVMs form an SVM ensemble; 4, performing integration on the models in the SVM ensemble by adopting an AdaBoost algorithm to obtain an integrated SVM model; and 5, for a given query sequence, generating eigenvectors corresponding to all residues in the sequence by using the same feature extraction method. For each residue sample, the integrated SVM model is used for performing prediction to generate an original prediction result, and then the original result is processed by utilizing a simple post-processing technology to generate a final prediction result. The method is high in prediction precision and good in generalization capability.

Description

technical field [0001] The invention relates to the field of bioinformatics protein-ligand interaction, in particular, it is a protein ligand binding site prediction method based on granular support vector machine integration. Background technique [0002] In the life activities of cells, proteins often need to bind with other molecules (ligands) to participate in various biological processes. Accurate identification of protein ligand binding sites is helpful for understanding protein function and designing new drugs. However, traditional biochemical identification methods are time-consuming and costly, and cannot meet the urgent needs of related research. Therefore, in recent decades, researchers in this field have proposed a large number of efficient computational methods to identify protein ligand binding sites, including: template-based methods, machine learning-based methods, etc. [0003] Machine learning-based methods are one of the most commonly used methods for pr...

AI Technical Summary

Problems solved by technology

[0004] However, in machine learning-based methods, the problem of class imbalance in datasets is an unavoidable problem, i.e., the number of ligand-binding residues (positive samples) is much smaller than that of non-ligand-binding residues (negative samples) Number of

However, blindly merging the current negative information granularity with all old negative information granularities can easily lead to redundancy of information

Images