CatGeoBench: Mechanism-Guided Catalytic Geometry for Enzyme Complex Co-Folding

Existing benchmarks for protein-ligand structure prediction and docking evaluate binding geometry and physical plausibility, but do not assess whether predicted enzyme complexes satisfy the geometric prerequisites of catalysis. We present CatGeoBench, an ongoing benchmark for catalytic geometric feasibility based on mechanisms from M-CSA. For each mechanism we build electron-flow chains and extract reacting atom pairs as the intermolecular bond-forming atom pairs collected up to the first structure-changing electron-flow event. Distance thresholds based on scaled van der Waals radii then provide quantitative criteria for these atom pairs. The current working set comprises 271 entries (369 reacting atom pairs), each with a curated enzyme functional unit, compositionally consistent substrate and cofactor, and explicit atom mappings to predicted coordinates. On crystal structures in which the same reacting atom pairs can be assigned, the observed distances are consistent with the proposed thresholds. We report preliminary co-folding results for Boltz-1x and Boltz-2; the model-agnostic protocol will support broader evaluation as the dataset is finalized. The pipeline can further extend to family-related enzymes that share mechanisms but act on alternative substrates.