Molecular generation method based on subgraph-variational self-encoding structure

Abstract

A molecular generation method based on a subgraph variation self-encoding structure comprises the steps: mining frequent subgraph sets of all molecules in a training set, and then screening frequent subgraphs through counting thresholds and pharmaceutical properties based on statistics to obtain a molecular subgraph total set S* meeting conditions; performing graph decomposition on each molecule Gin the training set to obtain a corresponding molecule sub-graph set S; inputting the molecular set S into an encoder for encoding to obtain a corresponding hidden space vector Z; and finally, inputting Z into a decoder for decoding to obtain a reconstructed molecule G. The method overcomes the defects of chemical property loss, difficulty in synthesis and the like caused by the traditional atom-by-atom generation method, ensures certain chemical properties of the molecule through the application of the substructure, and further generates a molecular structure with stable chemical propertiesand easiness in synthesis; the method brings great convenience to the discovery and design of novel drug small molecules, reduces the dependence of related workers on domain knowledge, improves the design and development efficiency of novel drugs, and shortens the research and development cycle.

Description

technical field [0001] The invention belongs to the field of artificial intelligence drug research and development, in particular to the field of drug research and development based on molecular subgraph-self-encoding structure molecular generation and optimization. Background technique [0002] In the field of drug molecule generation and optimization, due to the large number of molecular arrangements and combinations, the molecular space formed is too large, and the space for screening through artificial custom rules is limited. How to use existing molecular data to automatically find the most relevant target molecule? Molecules or molecular spaces are critical. Traditional methods usually analyze the chemical properties of known molecules and target molecules, and artificially design new molecular structures. This method is very inefficient, and this process is very dependent on the domain knowledge and experience of chemists, so some people propose a method based on mac...

AI Technical Summary

Problems solved by technology

Although the molecules generated by this graph-based method have better expressiveness than SMILES in some chemical properties, but because the entire graph is gradually generated based on a single atom, some global chemical properties are not considered in the generation process. structure, so the obtained molecules usually do not meet the actual needs in some chemical properties

Images