Deep learning prediction method of target-ligand binding affinity based on gated attention mechanism

Abstract

The invention relates to a deep learning prediction method of target-ligand binding affinity based on a gated attention mechanism, and belongs to the field of computer-aided drug design technology and biology and pharmacoinformatics. The deep learning model starts from an SMILES character string of a ligand and an amino acid sequence of a protein, and then is converted into a ligand matrix and a protein matrix. The ligand matrix is sent to a full connection layer and an attention layer based on gate enhancement for feature extraction, and the protein matrix is sent to a one-dimensional convolution layer and a maximum pooling layer and then sent to the attention layer based on gate enhancement. Finally, the processing characteristics of the ligand matrix are polymerized by adding matrix rows, and the same process is performed on the protein matrix, and then the two are spliced together and sent to a subsequent full connection layer to predict the probability of high / low binding affinity of the protein-ligand complex. According to the method, the time and cost related to experimental analysis are effectively reduced, and the efficiency of drug design and virtual screening is improved.

Description

technical field [0001] The invention relates to computer-aided drug design technology and the fields of biology and drug informatics, in particular to a target-ligand binding affinity prediction method. Background technique [0002] Most biological processes are determined by biomolecular recognition, where proteins often act as targets to interact with ligands to regulate biological functions, such as enzymatic catalysis, signal transduction, etc. The study of target-ligand interactions is an important topic. The quantity of binding strength (measured as a real number) of a target-ligand interaction is usually defined as binding affinity, which can be determined by the inhibition constant K i , the dissociation constant K d and half-maximal inhibitory concentration IC 50 to quantify. Most of the existing drugs are small molecule compounds with biological activity. Identifying ligands with high affinity to target proteins (small molecule drug candidates) is a major task ...

AI Technical Summary

Problems solved by technology

Based on data-driven methods (such as machine learning methods), the calculation speed is extremely fast, however, most machine learning algorithms still rely on expert knowledge for feature extraction / selection, and it is difficult to trade off accuracy and versatility in large data sets

Driven by massive data and powerful parallel computing capabilities, the deep learning method further developed from the traditional machine learning method has stronger data fitting capabilities. Although many deep learning models have been proposed to predict binding affinity, most Requires 3D structural features of the target-ligand complex, which are not readily available compared to textual features

Images