Gene-R1: Reasoning with Data-Augmented Lightweight LLMs for Gene Set Analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene Set Analysis (GSA) is crucial for identifying molecular functions of gene groups.
• Large Language Models (LLMs) are increasingly used for GSA, offering functional annotations and insights.
• Current LLM approaches for GSA predominantly use proprietary models, raising cost and data privacy concerns.

Purpose of the Study:

• To introduce Gene-R1, a novel framework for enhancing open-source LLMs in GSA.
• To equip lightweight, open-source LLMs with advanced reasoning capabilities for GSA.
• To address the lack of research on advanced reasoning strategies in LLM-based GSA.

Main Methods:

• Developed Gene-R1, a data-augmented learning framework.
• Integrated step-by-step reasoning capabilities into lightweight, open-source LLMs.
• Conducted experiments on 1,508 in-distribution and 106 out-of-distribution gene sets.

Main Results:

• Gene-R1 achieved substantial performance gains on in-distribution gene sets, matching commercial LLMs.
• On out-of-distribution gene sets, Gene-R1 demonstrated comparable performance to commercial and large-scale LLMs.
• The framework exhibited robust generalizability across diverse gene sources.

Conclusions:

• Gene-R1 effectively enhances open-source LLMs for GSA, providing competitive performance.
• The framework offers a cost-effective and privacy-preserving alternative to proprietary LLMs for gene function discovery.
• Gene-R1 shows promising generalizability, making it suitable for diverse biological datasets.