Protein annotation improves with compact tool interaction agents
ProtAgent: Post-Training Compact Protein Tool Agents with Interaction-Level Credit Assignment
Artificial Intelligence
Summary
Identifying protein functions often requires using many specialized bioinformatics tools. The authors introduced ProtAgent, a smaller smart system that learns how to pick and use these tools effectively by practicing with them after initial training. ProtAgent can decide which tool to use, interpret its outputs, and combine the results better than much larger existing models. This approach helps recover known protein data more accurately, although it doesn’t guarantee biological correctness by itself.
What this means in practice
- •For bioinformatics engineers: Build smarter protein annotation pipelines that learn to effectively call and integrate existing bioinformatics tools with less computing resources.
- •For software developers in biotechnology: Develop compact, adaptive agents to streamline biological data interpretation workflows, potentially improving tool integration and output analysis.
Authors
Yuxiang Wang, Baojian Zhou, Qifang Zhao, Keyue Jiang, Yanghua Xiao, Xiaoxiao Xu
Abstract
Protein annotation agents must choose suitable bioinformatics tools, interpret their outputs, and use the resulting evidence in the final answer. Most existing systems rely on frontier-scale models. We ask whether a compact model can instead learn this behavior through post-training with real tool interaction. We present ProtAgent, a two-stage pipeline that uses supervised fine-tuning (SFT) to learn the tool interface and standard on-policy REINFORCE to improve the interaction policy. Intermediate tool calls receive action rewards, and the final response receives a terminal answer reward. Step-level returns assign these rewards to the responses that produced them and to preceding actions. The episode-level baseline instead assigns one episode return to every action. On a fixed closed-loop protocol with real bioinformatics tools, ProtAgent-8B achieves 61.00% success under the annotation-coverage metric, compared with 41.60% for DeepSeek-V3.2-Thinking 671B and 60.80% for the strongest evaluated frontier system. In the three-seed comparison, mean success is 42.70% with episode-level returns and 61.40% with step-level returns under identity return normalization; the corresponding group-relative results are 36.50% and 58.20%. The metric measures recovery of reference annotations under this protocol; it does not by itself establish biological correctness or complete evidence grounding.