Economically Optimal Process Decision-Making Using ML Models 

Members: University of Wisconsin-Madison, Geno

Project dates: 2026 – present

This project will create a digital twin fermentation model and decision-making tool to that leverages machine learning to predict production outcomes after process perturbations that are typical for large-scale manufacturing and predict economically optimal responses in real time. This research will leverage well-established genome-scale metabolic models calibrated with multi-omic data. Bayesian experimental design – a machine learning-based methodology for sequential design of experiments – will be leveraged to maximize model accuracy with minimal experimental resources. A rapid and robust machine learning model will be trained using simulations from the mechanistic model to recommend optimal operator decisions in real-time. By integrating machine learning to guide experimental design and accelerate fermentation modeling and using metabolic modeling to predict fermentations beyond the scope of traditional models, this project will simultaneously predict fermentation performance at strain, unit operation, and process system scales. This tool will allow bioindustrial manufacturing companies to better predict their scale-up operations, resulting in fewer failed runs and saving money while they produce needed chemicals and materials. 

Funding source: U.S. Department of Defense and National Science Foundation (NSF)