Drug molecule screening method and system

Abstract

The invention provides a drug molecule screening method and system, and belongs to the technical field of biological information processing, and the method comprises the steps: obtaining to-be-screened drug molecules; obtaining a molecular orientation information descriptor based on the molecular formula of the drug molecule; wherein the molecular orientation information descriptor is composed of a plurality of one-dimensional vectors representing molecular bond characteristics of the drug molecules; and processing the molecular orientation information descriptor by using a pre-trained classification model to obtain a molecular property classification result. According to the method, molecules can be flexibly coded, cyclic collection of molecular information is avoided, and the accuracy of molecular description is improved; a convolutional neural network is combined with a self-attention module to process molecular descriptors, more and more comprehensive molecular information features are extracted, and the model classification capability is improved.

Description

technical field [0001] The invention relates to the technical field of biological information processing, in particular to a drug molecule screening method and system based on a molecular orientation information descriptor and a self-attention mechanism. Background technique [0002] Molecular property prediction is a chemical information processing technology. In recent years, the development of deep learning networks has greatly improved the effect of molecular property prediction. There are two types of models that have demonstrated good molecular description ability and produced good results in molecular performance prediction, one is neural networks applied to calculate molecular fingerprints or molecular descriptors made by experts, and the other is based on molecular graphs. Architecture to learn molecular features by graph convolutional neural networks. [0003] The representatives of molecular fingerprints are RDKit molecular fingerprints, Morgan molecular fingerpr...

AI Technical Summary

Problems solved by technology

[0004] Although the above two types of model methods perform well, they also have their own defects. For example, when the molecular fingerprint is fixed to describe the molecule, the work of collecting molecular information is not flexible; the molecular graph convolutional neural network mostly uses Molecular information is transmitted at the center of the atom, and the atomic information will be collected cyclically, resulting in duplication and redundancy of information, which in turn reduces the molecular description ability; molecular property classification networks mostly use feed-forward neural network (FFN), Random Forest (RF) Networks with relatively simple structures, such as networks, have insufficient ability to extract key information of molecules, and the prediction structure is not accurate enough.

Images