Molecular Representation Learning via Hierarchical Graph Transformer

Molecular Representation Learning (MRL) is widely applied in various downstream tasks, such as molecule generation, molecular property prediction and reaction prediction. Nevertheless, MRL faces several challenges posed by the vast chemical space and limited labeled-data availability. In this paper, we propose Hierarchical Graph Transformer (HieGT), integrating atom-level and motif-level representations to capture local-global characteristics of molecules over a hierarchical graph. Leveraging 2D topological and 3D geometric encoding, HieGT enhances intrinsic representation understanding of molecules. The proposed method achieves the state-of-the-art performance over the molecular property prediction dataset PCBA of Open Graph Benchmark (OGB), and competitive results on PCQM4Mv2 with better interpretability.