IAN, an intelligent system for omics data analysis and discovery.

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence in Genomics

Background:

• Gene expression data analysis is complex.
• Integrating diverse biological datasets (pathways, regulatory networks, protein-protein interactions) is challenging.
• Interpreting large-scale transcriptomic data requires advanced computational tools.

Purpose of the Study:

• To introduce IAN, an R package for artificial intelligence-driven gene expression data integration and interpretation.
• To leverage large language models (LLMs) for summarizing and explaining enrichment analysis results.
• To provide researchers with data-driven insights, system overviews, and key regulator identification.

Main Methods:

• Utilizes popular pathway databases (KEGG, WikiPathways, Reactome, GO) and regulatory datasets (ChEA).
• Integrates protein-protein interaction data from STRING.
• Employs a multi-agent architecture with LLMs for summarizing and interpreting enrichment analysis results.
• Applies engineered prompts and grounding instructions to guide LLM interpretation.

Main Results:

• Demonstrates IAN's functionality on published human transcriptomic data.
• Evaluates the robustness of IAN's performance across different LLMs.
• Presents human expert evaluations confirming the clarity and relevance of IAN's outputs.
• Provides explanations, system overviews, key regulators, and data-driven insights.

Conclusions:

• IAN offers a novel approach to gene expression data analysis by integrating AI and LLMs.
• The package facilitates the interpretation of complex transcriptomic data, enhancing biological discovery.
• IAN's modular design and expert-validated outputs make it a valuable tool for researchers.