Pyrite benchmarks LLM agents on protein-ligand pose ranking
Original title: Pyrite: Building and Benchmarking Protein-Ligand Pose Ranking Agents
On 1,500 pose pairs, PyriteFeaturizer reaches 74% accuracy and PyriteReasoner 67%, while expert chemists perform near chance.
Pyrite is a pairwise pose-ranking benchmark of 1,500 pose pairs across protein-ligand complexes. Expert chemists using 3D visualization and physics-based scores perform near chance. PyriteRep encodes geometry, interactions and physicochemical scores. Two LLM agents, PyriteReasoner and PyriteFeaturizer, reach up to 67% and 74% accuracy. LLM-guided feature discovery generally beats direct reasoning, and GRPO lets open-weight models match or surpass closed-source frontier models in the authors' evaluations.
- The benchmark has 1,500 pose pairs.
- PyriteReasoner reaches up to 67% accuracy via text comparison.
- PyriteFeaturizer reaches up to 74% by generating features for logistic regression.
- GRPO improves open-weight models to match or beat closed-source models in these tests.
From the original article
1 min read at arxiv.orgPredicting how a ligand binds to its target is central to small-molecule drug discovery. Modern cofolding models can generate diverse candidate poses, but reliably ranking these candidates remains challenging. We introduce Pyrite, a pairwise pose-ranking benchmark comprising 1,500 pose pairs across diverse protein-ligand complexes.
By Megha Srivastava · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.