RL-GCL: Reinforcement Learning-Guided Contrastive Learning for molecular property prediction

Yanran Fu, Junshan Han, Yong Xu, Kunhong Liu, Jiang Lu, Song He, Xiaochen Bo

Abstract: Highlights•RL generates hard molecular augmentations to enrich contrastive learning.•The introduction of labeling information reduces bias in false negative samples.•Two fine-tuning methods are incorporated: semi-supervised and linear protocol.•Bond and atom changes are analyzed to identify key substructures.