{
  "abstract": "Predicting protein functions from amino acid sequences is essential to understanding biological processes. Large language models (LLMs) offer flexible function descriptions to natural-language queries, yet their limited protein understanding motivates retrieval-augmented generation (RAG). Existing protein RAG methods primarily expand external retrieval signals while overlooking the rich knowledge encoded within a single protein language model (PLM). We show that retrieval performance varies across PLM layers and biological tasks, making conventional final-layer selection suboptimal. To this end, we propose QProtFuse, which uses queries to adaptively fuse multi-layer protein features and jointly learns protein and annotation retrieval through intra-modal and cross-modal contrastive objectives. On Prot-Inst-OOD, QProtFuse achieves the best protein and annotation retrieval scores, reaching 16.4\\% Recall@1 with annotation reranking. These gains improve downstream RAG-based function descriptions, yielding the highest LLM-as-a-judge scores for both Qwen3-8B (6.22/10) and DeepSeek-V4-Pro (6.68/10). Ablations and weight analyses demonstrate the value of query-guided fusion and complementary training objectives for exploiting a PLM's internal knowledge.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.30180",
  "author_total": 7,
  "authors": [
    {
      "name": "Mianzhi Pan"
    },
    {
      "name": "Peishuo Liu"
    },
    {
      "name": "Yawen Ouyang"
    },
    {
      "name": "Botian Wang"
    },
    {
      "name": "Yiming Rong"
    },
    {
      "name": "Hao Zhou"
    },
    {
      "name": "Jianbing Zhang"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/5jwwvjt6tzq4m.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.30180",
  "fulltext_url": null,
  "id": "5jwwvjt6tzq4m",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.30180",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "http://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.IR"
    ],
    [
      "arxiv",
      "q-bio.QM"
    ],
    [
      "openalex",
      "1312|Machine Learning in Bioinformatics"
    ]
  ],
  "openalex": "W7223341538",
  "pdf_url": "https://arxiv.org/pdf/2610.30180",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-19",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/5jwwvjt6tzq4m.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:34:38Z",
      "native_id": "2610.30180",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.30180"
    }
  ],
  "subjects": [
    {
      "field": "biology",
      "origin": "arxiv:q-bio.QM",
      "subfield": "bioinformatics"
    }
  ],
  "title": "QProtFuse: Query-Guided Layer Fusion for Retrieval-Augmented Protein Function Prediction",
  "title_text": "QProtFuse: Query-Guided Layer Fusion for Retrieval-Augmented Protein Function Prediction",
  "type": "preprint",
  "updated_at": "2026-10-30T12:34:38Z",
  "url": "https://research.draconitic.com/paper/5jwwvjt6tzq4m",
  "venue": "arXiv",
  "versions": []
}