An agentic AI environment to support biomedical research in collaborative academic environments

Agentic artificial intelligence (AAI) is being adopted by biomedical researchers to simplify the integration of computational, statistical, and data-science methods into their work. Despite this progress, several challenges persist, which cannot be overcome merely by deploying more powerful AI systems. Researchers need a technical understanding of how to use AAI, create standard operating procedures (SOPs) for data management, sharing, and curation, and establish rules to protect data privacy and security.

The commercial AAI systems currently available have been primarily designed for software engineering, necessitating substantial reconfiguration for biomedical data science applications. To address these issues, the authors introduce Murmurent, a shared software platform that operates beneath the agentic AI. Murmurent equips researchers with the necessary skills, procedures, and safeguards to effectively utilize AAI without the need for individual lab development.

Murmurent offers several key features. Firstly, it provides infrastructure to support collaborative projects and choreographies involving multiple lab members. Secondly, it features a set of specialized agents, each tailored to specific biomedical data science tasks such as software and statistical development, data visualization, literature search, and equity, diversity, inclusion, and decolonization (EDID) review.

Thirdly, it includes a tiered memory that can retain crucial research decisions, intermediate outputs, and sensitive data, enabling the system to automatically build more effective contexts in AAI sessions. Fourthly, Murmurent generates traceability records, aiding the AAI session in planning and executing data analyses and software builds more efficiently by avoiding repeating decisions that lead to dead-ends.

Fifthly, it enforces SOPs for data maintenance and governance across all lab members. Lastly, Murmurent supports multi-user interaction and collaboration among research groups.

To demonstrate Murmurent's capabilities, the authors employed it to identify potential inhibitors of Peptidyl-prolyl cis-trans Isomerase NIMA-interacting 1 (Pin1), a protein with a shallow catalytic site that makes it challenging to target. The authors describe the development of multiple approaches to identify Pin1 inhibitors and share the results obtained. Murmurent is open source, ensuring accessibility and transparency for the biomedical research community.