Reproduce “REBIND: Enhancing ground-state molecular conformation via force-based graph rewiring” 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
Taewon Kim, Hyunjin Seo, Sungsoo Ahn, Eunho Yang
REBIND introduces a deep learning framework that enhances 3D molecular conformation prediction by explicitly modeling inter-atomic forces, particularly non-bonded interactions, through a novel graph rewiring mechanism. This approach achieves up to a 24% reduction in conformational and energy-weighted RMSD on the QM9 dataset, generating more energetically stable and geometrically accurate structures.
Key insights
- Explicitly integrating physics-informed force modeling into graph neural networks significantly enhances the accuracy and physical realism of predicted molecular conformations.
- Dynamically rewiring graph structures based on calculated attractive and repulsive forces is crucial for capturing complex inter-atomic interactions beyond simple connectivity.
- Targeted augmentation of non-bonded edges, especially for low-degree atoms, effectively addresses a critical limitation in existing models and improves prediction reliability for energetically stable ground states.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.