The abstract is read at the publisher; the summary is JClub's.
StoiAgent predicts protein complex stoichiometry on 56 CASP17 targets with 75.0% top-1 accuracy, improving over PreStoi with AlphaFold3 ranking by 23.2 percentage points.
- Why it matters: Protein complex modeling depends on subunit copy numbers, which are often unknown, making stoichiometry prediction a major bottleneck for accurate structure modeling.
- What they did: StoiAgent uses the open-weight LLM Qwen3.6-27B to retrieve and integrate sequence, structural templates, AlphaFold3 predictions, and literature evidence for stoichiometry candidate generation and ranking.
- The result: It reaches 85.7% top-3 and 87.5% top-5 accuracy versus StoPred’s 80.4% and 85.7%, and it outperforms general-purpose frontier LLMs like ChatGPT and Gemini.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.