Protein multi-source feature fusion drug-target affinity prediction method

Abstract

The invention discloses a method for predicting drug-target affinity based on protein multi-source feature fusion, which comprises the following steps: firstly, constructing a PPI network and an SSN network, extracting protein features from the networks, and then collecting protein features such as subcellular positions, sequence codes, functional sites and structural domains for protein characterization; and fusing multi-source features by using a variational graph auto-encoder, and finally, inputting the multi-source features into a full-connection layer in combination with drug branches to carry out affinity prediction. According to the invention, a PPI network and an SSN network are constructed, so that biological priori knowledge between a target protein and other proteins is learned in addition to focusing on the characteristics of the target protein; according to the method, the protein characteristics are extracted and fused from the aspects of protein interaction, sequence similarity and protein subcellular positions for the first time, so that the drug-target affinity is predicted, and the prediction accuracy is improved; in addition, the characteristic source of the protein does not comprise a protein structure, so that the dependence on the protein structure is abandoned.

Description

technical field [0001] The invention relates to the technical field of biological information, in particular to a method for predicting drug-target affinity based on fusion of protein multi-source features. Background technique [0002] After the drug molecule enters the human body, it binds to the target protein, so as to produce drug effect and achieve the purpose of treating the disease. In the process of drug research and development, researchers will obtain a large number of candidate molecules, and it is very necessary to conduct virtual screening of these molecules. Predicting drug-target binding affinity is one of the important links, which indicates the relationship between drug molecules and target proteins. Bond strength. By predicting drug-target binding affinity to screen out molecules that do not meet affinity requirements, the search space for potential drug molecules is greatly reduced, eliminating the need for medical researchers to conduct extensive clinic...

AI Technical Summary

Problems solved by technology

[0006] The existing one-dimensional protein sequence-based and two-dimensional spatial structure-based affinity prediction models only extract features for the sequence or structure of a single target protein, while ignoring the rich biological prior knowledge between proteins, such as protein The interaction and homology information between them also ignore the expression of protein characteristics such as subcellular location and family, and the accuracy of affinity prediction needs to be further improved; in addition, the affinity prediction model based on two-dimensional structure still needs to rely on protein structures, while a large number of protein structures remain unresolved

Images