Molecular property prediction method based on molecular spatial position coding attention neural network model

Abstract

The invention discloses a molecular property prediction method based on a molecular spatial position coding attention neural network model. According to the method, spatial positions in a 3D conformation of a molecule are coded through a machine learning technology, so that the influence of different positions in the molecule on a substructure is better represented, the molecule is better represented, and meanwhile, a neural network structure of an attention mechanism and a gated loop network (GRU) are used for predicting chemical properties of the molecule. Herein, the topological relation of molecular substructures is fully utilized, experiments are carried out on a public data set, the effect of predicting the properties such as water solubility, toxicity and hydrophilicity of molecules is effectively improved, and thereby a new method is provided for drug molecule prediction.

Description

technical field [0001] The present invention relates to the fields of cheminformatics, machine learning / artificial intelligence, in particular to a theoretical method for predicting properties of unknown molecules by means of cheminformatics, machine learning / artificial intelligence based on spatial position coding. Background technique [0002] The drug R&D process has the characteristics of high capital density, high risk and long cycle, and requires a lot of capital, manpower and material resources. In the field of drug development, although many potential drug molecules have been extensively studied in animal models, there are still more than 30% of drug candidates that fail in practical applications because other intrinsic properties of the molecule do not meet the requirements. Property prediction work is of great value and can be used to better predict early molecular properties, which can greatly reduce the burden of later process failures, save a lot of resources an...

AI Technical Summary

Problems solved by technology

Secondly, using atoms directly as the basic unit of the graph is not conducive to maintaining molecular chemical information. Most existing graph networks use molecules as nodes for training, which often ignores the internal information between molecular substructures. Currently, based on sequences, chemical Molecules are regarded as sentence vectors. Through methods similar to natural language processing, representations of different molecules or the same substructure of the same molecule in different positions cannot be well distinguished, which reduces the performance of the model.

Images