benchmarks evals
The Binding Pose Becomes an Agent Test
A 1,500-pair benchmark asks models and chemists to rank candidate protein-ligand poses that cofolding systems already know how to produce.
Summary
A 1,500-pair benchmark asks models and chemists to rank candidate protein-ligand poses that cofolding systems already know how to produce.
Pyrite turns protein-ligand pose selection into 1,500 pairwise comparisons across diverse complexes. In the reported study, expert chemists using 3D viewers and physics-based scores performed near chance, while an LLM reasoner reached up to 67% accuracy and an agent that wrote features for logistic regression reached 74%. The authors found that LLM-guided feature discovery generally beat direct textual reasoning and that reinforcement learning improved open-weight models. The benchmark measures ranking within its constructed pairs; it does not establish clinical or drug-development performance.
Why it matters
A 1,500-pair benchmark asks models and chemists to rank candidate protein-ligand poses that cofolding systems already know how to produce.
Limits and context
- The benchmark measures ranking within its constructed pairs; it does not establish clinical or drug-development performance.
Key claims
A 1,500-pair benchmark asks models and chemists to rank candidate protein-ligand poses that cofolding systems already know how to produce.
Qualification: The benchmark measures ranking within its constructed pairs; it does not establish clinical or drug-development performance.
Evidence: source-2026-11-11-004
Sources
- arXiv:2611.14511v1arXiv / Cornell University · primary research
Corrections
No corrections have been recorded for this story.