Reproduce “Solvaformer: an SE(3)-equivariant graph transformer for small molecule solubility prediction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jonathan Broadbent, Michael Bailey, Mingxuan Li, Abhishek Paul, Louis De Lescure, Paul Chauvin, Lorenzo Kogler-Anele, Yasser Jangjou and 1 more
Solvaformer, an SE(3)-equivariant graph transformer developed by Sanofi, predicts small molecule solubility by combining geometric awareness with a unique dual-attention mechanism for multi-molecular systems. The model achieves near-quantum-mechanical accuracy with significantly faster inference, while also providing interpretable attention maps that elucidate key chemical interactions like hydrogen bonding.
Key insights
- Combining SE(3)-equivariant attention for fixed molecular structures with scalar attention for dynamic relative orientations of molecules effectively models solution-phase properties.
- Multi-task training, leveraging the consistency of quantum-mechanical data and the real-world relevance of experimental data, enhances model robustness and generalization capabilities.
- The attention mechanisms can provide chemically coherent interpretations, identifying crucial interactions such as differentiating between intramolecular and intermolecular hydrogen bonds that dictate solubility.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.