QProtFuse: Query-Guided Layer Fusion for Retrieval-Augmented Protein Function Prediction
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.
Related
- RA-PLA: Retrieval-augmented graph convolutional networks for protein-ligand binding affinity prediction
- Multi-vector retrieval enables residue-resolved prediction of protein partners by ColBERT-PPI
- EpiRank: partner-agnostic surface-patch retrieval for epitope prediction of protein binding interfaces
- GEO-BERT: A Retrieval augmented LLM Framework for Intelligent Dataset Retrieval from the Gene Expression Omnibus
- AltereGO: Prediction of Protein Function by Multimodal Fusion of Protein Language Models.
BibTeX
@misc{pan2026qprotfuse,
title = {{QProtFuse: Query-Guided Layer Fusion for Retrieval-Augmented Protein Function Prediction}},
author = {Mianzhi Pan and Peishuo Liu and Yawen Ouyang and Botian Wang and Yiming Rong and Hao Zhou and Jianbing Zhang},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.30180},
eprint = {2610.30180},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.30180},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - QProtFuse: Query-Guided Layer Fusion for Retrieval-Augmented Protein Function Prediction
AU - Mianzhi Pan
AU - Peishuo Liu
AU - Yawen Ouyang
AU - Botian Wang
AU - Yiming Rong
AU - Hao Zhou
AU - Jianbing Zhang
PY - 2026
DA - 2026/10/19
PB - arXiv
DO - 10.48550/arxiv.2610.30180
UR - https://arxiv.org/abs/2610.30180
AB - 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.
ER -