GenoMAS: A Multi-Agent Framework for Scientific Discovery via Code-Driven Gene Expression Analysis

2025-01-22 · Source: cs.MA updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Computational Biology & Genomics · Depth: Expert, extended

Summary

GenoMAS is a multi-agent framework designed to automate gene expression analysis, addressing challenges like complex data files and the need for extensive domain expertise. It integrates structured workflows with the adaptability of autonomous agents, orchestrating six specialized LLM agents through typed message-passing protocols. A guided-planning framework allows programming agents to convert high-level tasks into "Action Units" and dynamically adjust execution, including advancing, revising, bypassing, or backtracking. Evaluated on the GenoTEX benchmark, GenoMAS achieved an 89.13% Composite Similarity Correlation for data preprocessing and a 60.48% F1 score for gene identification, outperforming prior art by 10.61% and 16.85% respectively. The system also identifies biologically plausible gene-phenotype associations and adjusts for confounders, with its code publicly available.

Key takeaway

For bioinformatics and AI research scientists developing automated scientific analysis pipelines, GenoMAS demonstrates that integrating guided planning with a heterogeneous multi-agent LLM architecture significantly enhances performance and robustness. You should consider adopting similar principles of role specialization, iterative code generation/review, and dynamic context-aware planning to achieve higher accuracy and reduce computational costs in complex, domain-specific computational tasks, especially where precise procedural control and adaptive error handling are critical.

Key insights

GenoMAS automates gene expression analysis using a multi-agent LLM framework that balances structured workflows with adaptive, code-driven problem-solving.

Principles

• Combine structured workflows with autonomous agent adaptability.
• Utilize heterogeneous LLMs for complementary strengths.
• Implement iterative code generation, review, and revision.

Method

GenoMAS orchestrates six specialized LLM agents via typed message-passing and a guided-planning framework. Programming agents generate, revise, and validate executable code for Action Units, supported by Code Reviewer and Domain Expert agents.

In practice

• Use multi-agent systems for complex scientific automation.
• Employ guided planning to manage workflow flexibility.
• Integrate domain experts for biomedical reasoning in code generation.

Topics

• Gene Expression Analysis
• Multi-Agent Systems
• LLM Agents
• Code-Driven Scientific Discovery
• Genomic Data Preprocessing

Code references

• Liu-Hy/GenoMAS
• QwenLM/Qwen3

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• From Data to Theory: Autonomous Large Language Model Agents for Materials Science — Autonomous LLM agents can perform end-to-end scientific theory development, from equation selection to code execution and validation.
• 4 ways researchers are collaborating with Co-Scientist to solve big problems — Co-Scientist is a multi-agent AI system accelerating scientific hypothesis generation and refinement across life sciences.
• InfiniteScienceGym: An Unbounded, Procedurally-Generated Benchmark for Scientific Analysis — Procedurally generated benchmarks offer controlled, scalable evaluation for LLM scientific reasoning and abstention.
• Towards Generalist Agents for Accelerating Scientific Discovery171 — LLMs can accelerate scientific discovery by generating and refining hypotheses within agentic, iterative workflows.
• AutoNumerics: An Autonomous, PDE-Agnostic Multi-Agent Pipeline for Scientific Computing — AutoNumerics autonomously designs and verifies transparent PDE solvers from natural language using a multi-agent framework.
• Accelerating scientific discovery with Co-Scientist — Co-Scientist, a Gemini-based multi-agent AI, accelerates scientific discovery by generating and refining novel, verifiable hypotheses.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.MA updates on arXiv.org.