HAGO-Net: Hierarchical Geometric Message Passing for Molecular Representation Learning

• DOI: 10.1609/aaai.v38i13.29373
• Publication Date: 2024-03-25T11:23:02Z
• Authors: Hongbin Pei; Taile Chen; Chen A; Huiqi Deng; Jing Tao; Pinghui Wang; Xiaohong Guan
• Keywords: Net (polyhedron); Representation

Molecular representation learning has emerged as a game-changer at the intersection of AI and chemistry, with great potential in applications such drug design materials discovery. A substantial obstacle successfully applying molecular is difficulty effectively completely characterizing geometry, which not been well addressed to date. To overcome this challenge, we propose novel framework that features geometric graph, termed HAGO-Graph, specifically designed graph model, HAGO-Net. In framework, foundation enables complete characterization geometry hierarchical manner. Specifically, leverage concept n-body physics characterize patterns multiple spatial scales. We then message passing scheme, HAGO-MPS, implement scheme neural network, HAGO-Net, learn HAGO-Graph by horizontal vertical aggregation. further prove DHAGO-Net, derivative function an equivariant model. The proposed models are validated extensive comparisons on four challenging benchmarks. Notably, exhibited state-of-the-art performance chirality identification property prediction, achieving five properties QM9 dataset. also achieved competitive results dynamics prediction task.