Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Single to Multi-Agent Reasoning: Advancing GeneGPT for Genomics QA

Kimia Abedini, Farzad Shami, Gianmaria Silvello

TL;DR

This paper tackles genomic QA limitations in large language models by introducing GenomAgent, a multi-agent framework that coordinates specialized agents to retrieve and extract data from diverse genomic databases. Building on GeneGPT's single-agent, tool-augmented approach, GenomAgent enables parallel API querying, dynamic data extraction, and consensus-based answer synthesis, significantly improving performance and efficiency. On the GeneTuring benchmark, GenomAgent achieves an average score of 0.93 (vs. 0.83 for GeneGPT) and reduces total cost to 2.11 (from 10.06), with especially large gains in sequence alignment (28.8%). This work demonstrates that coordinated multi-agent orchestration can provide superior genomics QA with substantial resource savings and suggests broader applicability to other scientific domains requiring expert knowledge extraction from multiple sources.

Abstract

Comprehending genomic information is essential for biomedical research, yet extracting data from complex distributed databases remains challenging. Large language models (LLMs) offer potential for genomic Question Answering (QA) but face limitations due to restricted access to domain-specific databases. GeneGPT is the current state-of-the-art system that enhances LLMs by utilizing specialized API calls, though it is constrained by rigid API dependencies and limited adaptability. We replicate GeneGPT and propose GenomAgent, a multi-agent framework that efficiently coordinates specialized agents for complex genomics queries. Evaluated on nine tasks from the GeneTuring benchmark, GenomAgent outperforms GeneGPT by 12% on average, and its flexible architecture extends beyond genomics to various scientific domains needing expert knowledge extraction.

From Single to Multi-Agent Reasoning: Advancing GeneGPT for Genomics QA

TL;DR

This paper tackles genomic QA limitations in large language models by introducing GenomAgent, a multi-agent framework that coordinates specialized agents to retrieve and extract data from diverse genomic databases. Building on GeneGPT's single-agent, tool-augmented approach, GenomAgent enables parallel API querying, dynamic data extraction, and consensus-based answer synthesis, significantly improving performance and efficiency. On the GeneTuring benchmark, GenomAgent achieves an average score of 0.93 (vs. 0.83 for GeneGPT) and reduces total cost to 2.11 (from 10.06), with especially large gains in sequence alignment (28.8%). This work demonstrates that coordinated multi-agent orchestration can provide superior genomics QA with substantial resource savings and suggests broader applicability to other scientific domains requiring expert knowledge extraction from multiple sources.

Abstract

Comprehending genomic information is essential for biomedical research, yet extracting data from complex distributed databases remains challenging. Large language models (LLMs) offer potential for genomic Question Answering (QA) but face limitations due to restricted access to domain-specific databases. GeneGPT is the current state-of-the-art system that enhances LLMs by utilizing specialized API calls, though it is constrained by rigid API dependencies and limited adaptability. We replicate GeneGPT and propose GenomAgent, a multi-agent framework that efficiently coordinates specialized agents for complex genomics queries. Evaluated on nine tasks from the GeneTuring benchmark, GenomAgent outperforms GeneGPT by 12% on average, and its flexible architecture extends beyond genomics to various scientific domains needing expert knowledge extraction.
Paper Structure (6 sections, 2 figures, 2 tables)

This paper contains 6 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. GeneGPT
  3. Reproducibility of GeneGPT
  4. GenomAgent
  5. Experiments
  6. Final Remarks and Future Work

Figures (2)

  • Figure 1: GenomAgent multi-agent architecture and workflow.
  • Figure 2: Performance-cost tradeoff on GeneTuring. Bubble size shows normalized cost; High Value Region shows optimal performance at minimal cost.