Automatic biomarker discovery and enrichment with BRAD.

Area of Science:

• Biomedical Research
• Computational Biology
• Artificial Intelligence

Background:

• Integrating Large Language Models (LLMs) into biomedical research tools faces technical and reproducibility hurdles.
• Commercial AI systems lack transparency and can produce inaccurate information, limiting their research utility.
• The Bioinformatics Retrieval Augmented Digital (BRAD) agent system was developed to overcome these limitations.

Purpose of the Study:

• To introduce BRAD, an agentic system designed to integrate LLMs with external tools and data for streamlined research.
• To enhance the reliability of AI-generated results through transparent protocols and modular agents.
• To automate complex research workflows, such as biomarker discovery and enrichment report generation.

Main Methods:

• BRAD employs modular agents to retrieve information from diverse sources, including scientific literature, custom software, and online databases.
• The system maintains transparent protocols to ensure data provenance and the reproducibility of AI-driven analyses.
• BRAD was applied to a biomarker discovery pipeline, automating execution and report generation.

Main Results:

• BRAD successfully automated a biomarker discovery pipeline, including data analysis and enrichment report generation.
• The system contextualizes user data within existing literature, offering enhanced interpretation capabilities.
• BRAD demonstrated flexibility through deployments in other applications like chatbots and single-cell data analysis.

Conclusions:

• BRAD offers a reliable and transparent solution for integrating LLMs into biomedical research workflows.
• The system enhances automation and interpretation, surpassing conventional research tools.
• BRAD provides a flexible platform for diverse AI-driven research applications.