Molecular feature extraction and performance prediction method based on image convolution

Abstract

The invention provides a molecular feature extraction and performance prediction method based on image convolution, and the method comprises the steps: carrying out the quantification of the information of atoms and chemical bonds between the atoms, forming a node feature matrix of a molecular image, extracting the connection information between the atoms in a molecule, and forming an adjacent matrix of the image, and fusing the feature matrix and the adjacent matrix into a network model based on image convolution to obtain a molecular feature matrix containing relatively complete atomic information, chemical bond information and molecular structure information, and then performing model training to obtain a final network model. According to the method, the molecular information is effectively captured, and the prediction precision of the model molecular performance is improved.

Description

technical field [0001] The invention relates to the technical field of molecular fingerprint design, in particular to a molecular feature extraction and performance prediction method based on image convolution. Background technique [0002] The prediction of molecular properties is the key to effective materials discovery and is an important part of materials genome research. With the improvement of computing power and the continuous development of molecular databases, machine learning has been widely used in chemistry and materials research, such as electronic structure learning, spectral property prediction, and virtual screening of related material design. The quantitative structure-activity relationship can be established more accurately and effectively. [0003] At present, molecular fingerprint design and appropriate molecular representation construction are a challenge for molecular machine learning. Molecular feature extraction is an important part of machine learni...

AI Technical Summary

Problems solved by technology

[0004] The traditional ECFP circular fingerprint uses the hash algorithm to encrypt the molecular substructure, turning it into a binary vector, but there is a problem of information loss during the encryption process; the CM Coulomb fingerprint uses the atomic charge and the distance between atoms to construct the Coulomb matrix, but the Fingerprints are not atomic number permutation invariant

Images