Protein sequence data improves gene set annotation with language models
SoftGene: Protein Language Model-Enhanced Soft Prompting for Interpretable Gene Set Annotation
Artificial Intelligence
Summary
Understanding groups of genes and their functions is a key challenge in biology that usually requires expert knowledge and lots of work. This paper presents SoftGene, a new way to teach computers to better explain gene groups by using information from the proteins these genes make. The method combines detailed protein data with smart language models to give clearer and more accurate descriptions of gene sets. The authors tested their approach on well-known biology databases and found it works better than older methods, although its benefits vary depending on the type of biological information.
What this means in practice
- •For bioinformatics teams: Automate clearer gene set functional annotation by integrating protein sequence data with language models to reduce manual curation effort.
- •For computational biology software developers: Build tools that combine protein sequence embeddings and hybrid language model prompts for improved gene function characterization.
Authors
Drew Ross, Arya Hadizadeh Moghaddam, Dongjie Wang, Xiaoyu Zhang, Zijun Yao
Abstract
Gene set analysis is a cornerstone of functional genomics, yet it remains labor-intensive and heavily dependent on manual curation and expert biological interpretation. While Large Language Models (LLMs) have emerged as powerful tools for genomic reasoning and annotation, most existing approaches rely on symbolic gene names and fail to capture domain-specific biological structure, particularly protein sequence information that governs molecular activity, interactions, and downstream gene function. In this work, we propose SoftGene, a novel framework for LLM-based gene set annotation that leverages the hierarchical structure of gene sets. First, we use a hierarchical attention-based encoder built on ESM, a protein language model, to represent each gene set using protein-level amino acid sequence information. Second, we construct a hybrid prompting scheme that combines soft prompts derived from gene set embeddings with hard prompts containing auxiliary context generated by an LLM, and feed the resulting prompt into a local LLM for annotation. We evaluate our framework on two benchmark datasets: Gene Ontology (GO) and the Molecular Signatures Database (MSigDB). Our results show that integrating protein-sequence representations with textual context improves gene set annotation overall, while per-domain analyses reveal that the contribution of protein embeddings varies across biological domains.