Graph Representation Learning for Drug Discovery
Artificial intelligence has seen big opportunities in drug discovery thanks to the large amount of data collected in this domain. Most of the data in the domain are represented as graph structures such as drug-protein interaction graph, protein-protein interaction graph, and molecular graphs. In this talk, I will introduce our recent work on developing deep learning and reinforcement learning techniques for graph representation learning and generation and their applications to drug discovery. Specifically, I will talk about (1) how to learn effective molecular graph representations for molecule properties prediction in unsupervised and supervised fashion; (2) A new generative model for molecular graph generation and optimization, which is able to generate 100% chemical valid molecules and meanwhile achieves state-of-the-art performance in both chemical properties optimization and constrained property optimization.