{
  "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.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.33746",
  "author_total": 6,
  "authors": [
    {
      "name": "Yuxiang Wang"
    },
    {
      "name": "Baojian Zhou"
    },
    {
      "name": "Qifang Zhao"
    },
    {
      "name": "Keyue Jiang"
    },
    {
      "name": "Yanghua Xiao"
    },
    {
      "name": "Xiaoxiao Xu"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/fj4j71wkxp1jc.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.33746",
  "fulltext_url": null,
  "id": "fj4j71wkxp1jc",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.33746",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.AI"
    ],
    [
      "arxiv",
      "q-bio.QM"
    ],
    [
      "openalex",
      "1312|Machine Learning in Bioinformatics"
    ]
  ],
  "openalex": "W7223794867",
  "pdf_url": "https://arxiv.org/pdf/2610.33746",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-24",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/fj4j71wkxp1jc.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:35:44Z",
      "native_id": "2610.33746",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.33746"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "arxiv:q-bio.QM",
      "subfield": "bioinformatics"
    }
  ],
  "title": "ProtAgent: Post-Training Compact Protein Tool Agents with Interaction-Level Credit Assignment",
  "title_text": "ProtAgent: Post-Training Compact Protein Tool Agents with Interaction-Level Credit Assignment",
  "type": "preprint",
  "updated_at": "2026-10-30T12:35:44Z",
  "url": "https://research.draconitic.com/paper/fj4j71wkxp1jc",
  "venue": "arXiv",
  "versions": []
}