Benchmark shows static refolding overstates motif-scaffold quality; MD checks expose gaps
- BILBY-PRO is a benchmark of conserved functional motifs drawn systematically from PROSITE-linked structures, used to compare six recent scaffold generators on target coverage and unique-solution yield.
- BILBY-MD adds all-atom simulations against references, tracking motif and scaffold drift, contacts and dihedral landscapes.
- Many statically solved MotifBench designs fail the MD criteria, especially context-dependent motifs, so refolding success alone does not show dynamical fidelity.
💡 Points Based on the abstract
Whether a design "fails" MD hinges on simulation length, replicate count and force field, so I'd want failure rates to hold across seeds before trusting any ranking of generators. Still, on my own project a refolding hit is only a first filter: a motif that drifts early in unrestrained MD rarely keeps the precise geometry that function needs, and I'd rather learn that before ordering genes.
Static Recovery Is Not Dynamic Stability: Dynamics-Aware Benchmarking of Protein Motif Scaffolding ↗