Molformer: Motif-Based Transformer on 3D Heterogeneous Molecular Graphs

Procuring expressive molecular representations underpins AI-driven molecule design and scientific discovery. The research mainly focuses on atom-level homogeneous graphs, ignoring the rich information in subgraphs or motifs. However, it has been widely accepted that substructures play a dominant role identifying determining properties. To address such issues, we formulate heterogeneous graphs (HMGs) introduce novel architecture to exploit both motifs 3D geometry. Precisely, extract functional groups as for small molecules employ reinforcement learning adaptively select quaternary amino acids motif candidates proteins. Then HMGs are constructed with motif-level nodes. better accommodate those HMGs, variant of Transformer named Molformer, which adopts self-attention layer distinguish interactions between multi-level Besides, is also coupled multi-scale mechanism capture fine-grained local patterns increasing contextual scales. An attentive farthest point sampling algorithm proposed obtain representations. We validate Molformer across broad range domains, including quantum chemistry, physiology, biophysics. Extensive experiments show outperforms achieves comparable performance several state-of-the-art baselines. Our work provides promising way utilize informative from perspective graph construction. code available at https://github.com/smiles724/Molformer.