Molecular Contrastive Learning with Chemical Element Knowledge Graph

• DOI: 10.1609/aaai.v36i4.20313
• Publication Date: 2022-07-04T11:00:38Z
• Authors: Yin Fang; Qiang Zhang; Haihong Yang; Xiang Zhuang; Shumin Deng; Wen Zhang; Ming Qin; Zhuo Chen; Xiaohui Fan; Huajun Chen
• Keywords: Molecular graph; ENCODE; Feature Learning; Knowledge graph

Molecular representation learning contributes to multiple downstream tasks such as molecular property prediction and drug design. To properly represent molecules, graph contrastive is a promising paradigm it utilizes self-supervision signals has no requirements for human annotations. However, prior works fail incorporate fundamental domain knowledge into semantics thus ignore the correlations between atoms that have common attributes but are not directly connected by bonds. address these issues, we construct Chemical Element Knowledge Graph (KG) summarize microscopic associations elements propose novel Knowledge-enhanced Contrastive Learning (KCL) framework learning. KCL consists of three modules. The first module, knowledge-guided augmentation, augments original based on KG. second knowledge-aware representation, extracts representations with encoder Knowledge-aware Message Passing Neural Network (KMPNN) encode complex information in augmented graph. final module objective, where maximize agreement two views graphs. Extensive experiments demonstrated obtained superior performances against state-of-the-art baselines eight datasets. Visualization interpret what learned from