Pretrained Graph Transformers for Chirality-related Tasks
Submission Track: Short Paper
Submission Category: AI-Guided Design
Keywords: pretrain, GNN, chirality, property prediction
Abstract: Graph Neural Networks (GNNs) play a fundamental role in cheminformatics. However, typical GNNs cannot capture the concept of chirality, which means they cannot distinguish chiral molecules from their enantiomers. The ability to distinguish between enantiomers is important as enantiomers may exhibit very distinct biochemical properties. Herein, we proposed a method that leverages the spatial structure of molecules to incorporate conformational information into GNNs. A Pretrained Graph Transformer was designed for Chiral tasks (ChiPGT), which can iteratively optimize raw 3D enantiomer conformations generated by inexpensive methods such as RDKit. The results indicated that our ChiPGT outperforms current state-of-the-art GNNs in chiral-sensitive molecular property prediction tasks on the D4DCHP and REDS datasets.
Submission Number: 1
Loading