A Molecular Structure Prediction Method Based on Graph Convolutional Networks

Abstract

The invention belongs to the technical field of machine learning, and in particular relates to a molecular structure prediction method based on a graph convolutional network. The present invention constructs a molecular graph and a molecular complete graph according to the SMILES of the input molecule, and correspondingly constructs a network model with two branches, one branch uses MLP for edge prediction, and the other branch includes graph convolutional network and MLP for The overall structural features of the branches are extracted. The invention uses graph convolution to extract molecular structure features, which can well extract the overall structural features of molecules, so as to better predict the structure; the double-branch model design is used to solve the problem that the complete graph destroys molecular structure information .

Description

Technical field [0001] The present invention belongs to the field of machine learning techniques, specifically relates to a molecular structure prediction method based on a graph convolutional network. Background [0002] The structure of the molecule is the basis for the study of the molecule, because the microstructure of the molecule is inseparable from the relationship between various macroscopic chemical properties including chemical reactions and various physical properties, so people have begun to study the molecular structure from a long time ago, and the regularity of the structure of the material structure is summarized from the molecular configuration and motion characteristics of various known chemical substances to explain the formation of the substance and its properties. [0003] The earliest molecular structure determination is mainly through experimental observation methods, such as microwave spectroscopy, X-ray diffraction, electron diffraction and neutron diff...

AI Technical Summary

Problems solved by technology

Among them, the molecular structure in the QM9 data set has been optimized by multiple methods and has high reliability, but it needs to switch between several methods, which is cumbersome and time-consuming. Although the latter methods are faster in calculation, they are often compared. rough or requires prior knowledge

[0005] In recent years, with the development of machine learning, more and more machine learning methods have been applied to the fields of materials and chemistry and have achieved success, which greatly reflects the advantages of machine learning methods in quantum chemical calculations. The emerging graph neural network is very suitable for feature extraction of molecular graph representations, but existing methods usually use it to predict the properties of molecules rather than structures. For example, Chen et al. developed a graph called MEGNet The network model accurately predicts the properties of molecules and crystals. Louis et al. proposed a new GATGNN model to solve the problem that many GNN models cannot effectively distinguish the contributions of different atoms when predicting properties, and achieved quite good results. predictive performance

However, the use of graph neural networks to predict molecular structures has not yet seen relevant public reports.

[0006] Because the molecular structure plays a very important role in the research, understanding and synthesis of molecules, especially in the field of new drug development, knowing the molecular structure of drugs is very helpful for the prediction of drug molecular properties, target prediction and chemical reaction prediction, so as to guide the synthesis of new molecules , and the determination of molecular structure by the experimental method is complicated, costly, and requires manpower and material resources, and it is impossible to determine the newly designed molecules that have not yet existed. In the traditional calculation method, there is a problem that the calculation accuracy and time cost cannot be well balanced. problem of relationship

Images