Symposium 2025 Recordings

In-silico comparison of classical response surface and hybrid pareto optimization for feed optimization on a ground-truth dynamic metabolic model

Dr. Amos Lu

Abstract

Accelerating development timelines necessitates more efficient experiment designs and utilization of data. We first developed an in-silico ground-truth model of a media/feed optimization task with a cellular dynamic metabolic model and a batch sequential experimental format mirroring high-throughput bioreactor operations.

Using the ground-truth model, we compare classical design of experiments approach (fractional factorial, quadratic response surface) against hybrid model-based design of experiments (Latin hypercube, pareto optimization) on the metrics of maximizing ultimate titer and minimizing the required number of experiments.

Lastly, we implement realistic uncertainties for analytical variability and contamination risks and evaluate their effects of experimental design performance.