zPocket wins Best Paper at MoML2026: frozen co-folding features beat pocket finders
edunov · x · 2026-10-17
zPocket, by MIT PhD student and Genesis Molecular AI intern Tally Portnoi, won the Best Paper prize at MoML2026. Key insight: co-folding models already encode where small molecules bind from protein-ligand complexes. Training a tiny head on frozen co-folding features yields per-residue pocket predictions that beat all compared pocket finders (UniSite-1D, P2Rank, VN-EGNN) on every stratum of the PocketWatch benchmark, including pockets most dissimilar to pre-cutoff records. The authors also flag that existing pocket benchmarks predate modern co-folding training data.
More from Research
- LeanGraphSearch mines Mathlib dependency graphs, lifting proof success to 81% — kaiwei_chang · 2026-10-17
- Kornia ships open-source Rust SLAM pipeline, seeking battle testers — kornia_foss · 2026-10-17
- Perplexity decider v1.2 lands #3 on Decision Index as longer training erodes backbone skills — antoine_chaffin · 2026-10-17
- Perplexity traces decider v1.2 regressions to longer training, mitigates via model souping — antoine_chaffin · 2026-10-17
- Packing multiple questions into one forward pass saves compute but adds instability — antoine_chaffin · 2026-10-17
- pplx-decider packs multiple questions per state into one bidirectional forward pass — antoine_chaffin · 2026-10-17