SciAgent OLab: Open Multi-Agent LLMs for Accelerating Biomedical Discovery

Suk Yee Yong, Robert Dunne, Carol Lee, Alex McAuley

Keywords: Agentic AI, Large Language Mode, Multi-Agent, Open-weight Models, Influenza, SARS-CoV-2 Co-infection

TL;DR: We present SciAgent OLab, a locally deployable framework that builds on Virtual Lab for biomedical research, using open-weight LLMs and multi-agent collaboration to automate literature-driven workflows securely and cost-effectively.

Abstract: Large language models (LLMs) combined with agentic AI frameworks are increasingly applied to accelerate biomedical research, offering new opportunities for generating insights that can guide real-world experimental studies. Existing tools such as Virtual Lab exemplify this approach but rely on proprietary, cloud-hosted LLMs, creating barriers related to privacy and cost. We present Science Agent Open Lab (SciAgent OLab), a framework that delivers Virtual Lab functionality using locally deployed, open-weight LLMs while maintaining multi-agent collaboration. SciAgent OLab coordinates role-specialized agents to automate literature-driven workflows, including query generation, document retrieval, summarization, and structured knowledge extraction, with optional human oversight for high-level guidance. To demonstrate its utility, we conducted a pilot case study on influenza and SARS-CoV-2 co-infection, where agents curated relevant publications and synthesized key insights to support research exploration. By enabling secure and cost-effective AI-driven workflows without reliance on proprietary services, SciAgent OLab provides a practical framework for accessible and scalable applications in biomedical research.

Submission Number: 2