Ligand compound rapid pre-screening model based on deep learning

Abstract

The invention provides a ligand compound rapid pre-screening model based on deep learning. The method includes the following steps: S1, constructing a data set, and encoding amino acid sequence information of binding sites of ligand compounds and receptor proteins; s2, combining the ligand and receptor structure vectors and inputting them into a deep neural network, outputting affinity scores, and training the model in a supervised learning mode; s3, sorting according to affinity scoring results, screening out positive compounds, and removing a large number of negative compounds; s4, obtaining molecular fingerprints (molecular characteristics) of positive compounds, and clustering based on a similarity or distance measurement method; and S5, taking the compound with the highest score in each cluster as a potential active compound. According to the method, rapid pre-screening of the ligand compound is achieved, redundancy can be removed through clustering analysis of the positive compound, the structural diversity of the ligand compound is guaranteed, and the prediction speed and accuracy are improved.

Description

technical field [0001] The invention belongs to the field of computer application technology, and in particular relates to a rapid pre-screening model for ligand compounds based on deep learning. Background technique [0002] As the starting point of drug development, the discovery of hit compounds is crucial to the entire drug development process. As an important technology for discovering hit compounds, virtual screening is the core technology of computer-aided drug design, that is, to quickly screen a small number of compounds from a large number of molecular databases. For target compounds, the pre-screening of ligand compounds can increase the pace of scientific research, reduce the number of compounds entering the biochemical experiment stage, and effectively improve the success rate and efficiency of drug development. [0003] The most commonly used method for the pre-screening of ligand compounds is molecular docking, through which the protein-ligand binding mode can...

AI Technical Summary

Problems solved by technology

[0003] The most commonly used method for the pre-screening of ligand compounds is molecular docking, through which the protein-ligand binding mode can be simulated, the ligand binding conformation can be predicted, and the affinity can be scored to determine potential active compounds. However, existing molecular The docking theory itself is not perfect. Molecular docking has low accuracy in predicting protein-small molecule binding mode and affinity. In addition, there are many docking programs to choose from, and their performances are different, resulting in a wide range of success rates in actual use.

Images