Property Prediction of Molecules in Graph Convolutional Neural Network Expansion

The prediction of molecular properties is further studies of molecular structure similarity and plays an important role in the drug development process. Graph convolutional neural networks provide an effective method to study the relationship between molecular structure and properties. Many variants...

Full description

Saved in:
Bibliographic Details
Published in2019 IEEE 10th International Conference on Software Engineering and Service Science (ICSESS) pp. 263 - 266
Main Authors Meng, Mei, Wei, Zhiqiang, Li, Zhen, Jiang, Mingjian, Bian, Yujie
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.10.2019
Subjects
atomic group
Atomic measurements
attention mechanism
Chemicals
Convolution
Lips
Logic gates
molecular graph
skip connection
Task analysis
Training
Online AccessGet full text

Cover

More Information
Summary:The prediction of molecular properties is further studies of molecular structure similarity and plays an important role in the drug development process. Graph convolutional neural networks provide an effective method to study the relationship between molecular structure and properties. Many variants of graph convolutional networks that have emerged in recent years have improved network performance from different perspectives. The expansion of the graph convolution neural network (ExGCN) is proposed in this paper. The graph convolutional neural network is expanded by incorporating the core ideas of the latest variants including Graph Attention Network, Gated Graph Neutral Network, and construct new molecular graphs. Though experiments on the toxicity data set of tox21, the lipophilicity data set of Lip, and the solubility data set of ESOL. The proposed method performs better compared to other traditional methods.
ISSN:2327-0594
DOI:10.1109/ICSESS47205.2019.9040723