AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-26
AtomMOF: All-Atom Flow Matching for MOF-Adsorbate Structure Prediction
AtomMOF tackles predicting how adsorbates bind inside metal-organic frameworks (MOFs), a task that is costly with traditional methods. It is an all-atom flow-matching model that jointly predicts MOF-adsorbate structures from building blocks, and guidance from machine-learned potentials improves structural validity. On BW it raised match rate by 58.17% relative and cut RMSD by 31.84%, while on ODAC25 it generated adsorption configurations faster than GCMC and found lower-energy configurations than the reference dataset after relaxation.