Molecular Property Prediction Based on Graph Contrastive Learning
Yanchun Jin, Xiaoying Winston Yan, Qing Li · 2024
Molecular Property Prediction (MPP) aids in drug discovery by predicting diverse molecular properties. This article presents a graph contrastive learning-based method for MPP. The approach effectively integrates substructure nodes into molecular graphs, enabling comprehensive molecular characterization. By combining specialized Graph Isomorphism Network (GIN) with contrastive learning, the method aims to improve prediction accuracy and generalization. The proposed method's effectiveness is confirmed by experimental results conducted on molecular property datasets.