To overcome manual, slow, and intuition-dependent traditional bioprocess optimization, this project introduces a modular, agentic AI platform designed to close the physical-to-digital-to-physical loop. Deployed at the Bakery Square AI Science Foundry, the system leverages Anthropic’s Claude Code to formulate hypotheses, design experiments, and generate automated robotic liquid handling protocols. Over a 1-year timeline, the platform will screen over 100 conditions per day using a BioLector XT to track yeast biomass and lycopene expression, targeting a >50% yield improvement with minimal human intervention. Data streams will feed into the PI's jaxsr (JAX Symbolic Regression) tool to discover sparse, interpretable mathematical formulas with calibrated uncertainty quantification, avoiding uninspectable black-box surrogates. All operational data flows will integrate into the MFI Digital Data Backbone (DDB) using FAIR-compliant metadata schemas, establishing a validated, domain-agnostic automation framework readily translatable to other biomanufacturing targets and cross-disciplinary applications like the Materials Innovation Cloud Lab (MICL) alloy discovery workflows.
Application/demonstration
The project will demonstrate an automated, closed-loop optimization campaign screening over 100 yeast cultivation conditions per day in a BioLector XT to achieve a greater than 50% increase in lycopene yield with human-in-the-loop dashboard oversight.