Modeling Integration and Data Agents for Science

EMSL is spearheading the development of an extensible Modeling, Integration, and Data Agents for Science (MIDAS) capability to support the Biological and Environmental Research (BER) community. 

MIDAS will constitute an ecosystem of interoperable artificial intelligence (AI)-based agents that will accelerate basic science research in the BER mission space by increasing the efficiency and effectiveness of researchers. MIDAS capabilities will also advance American scientific leadership and manufacturing competitiveness by creating essential AI models and software to enable and stimulate growth in the bioeconomy. 

AI-based agents represent the next paradigm for AI research and applications. In particular, by using large language models (similar to ChatGPT) at the intersection between human users and software and by allowing such models to suggest and execute the next steps of computational work, agents open transformative possibilities as “force multipliers” to accelerate research and collaboration between experts with different areas of specialization. 

EMSL’s unique position for creating critical new capabilities for BER science and the Nation begins with our exceptional breadth of experimental capabilities and staff expertise, including expertise with the analysis of the associated data types. As a user facility with decades of experience integrating those disparate capabilities to advance user science, EMSL is well positioned to identify high-value opportunities where agentic workflows in predictive modeling, data analytics, and integration can accelerate science by increasing efficiency for multidisciplinary research teams. 

Furthermore, MIDAS capabilities will complement EMSL’s lab automation and support the transition to autonomous science workflows. MIDAS agents will leverage EMSL’s new state-of-the-art laboratory information management system (LIMS) which captures complete, high-quality metadata regarding experimental samples. The LIMS enables MIDAS by ensuring that EMSL generates AI-ready data automatically, so that EMSL’s comprehensive data models can enable cutting-edge data integration between EMSL data and other BER data resources. 

To ensure success in delivering MIDAS, intramural investments and external partnerships will prioritize research on specializing models and agents for BER science, mechanisms for interagent AI collaboration enabling autonomous science, accelerating the design–build–test–learn (DBTL) loop, human–AI teaming, and BER-specific algorithms and data foundations.