Poster
in
Workshop: AI4MAT-ICLR-2025: AI for Accelerated Materials Design
MatDock: Multi-molecule docking in porous materials with flow matching
Malte Franke · Mingrou Xie · Akshay Subramanian · Juno Nam · Rafael Gomez-Bombarelli
Keywords: [ flow matching ] [ Molecular docking ] [ generative models for materials ] [ porous materials ]
Molecular docking in materials is important for creating geometries for downstream computations such as structure optimization and transition-state finding. In this work, we present the first use of generative models for multi-molecule docking in periodic materials. MatDock uses flow matching to dock multiple molecules of the same identity in periodic materials. We illustrate its use in docking molecules in porous materials (zeolites) and compare between uniform sampling and Voronoi-based sampling methods. MatDock can be extended beyond just docking to generating energy-optimized docked structures, thus bypassing the key computational bottleneck in creating material-molecule complexes.