Uncertainty-Driven Experiment Design in Cell-Free Protein Synthesis with Bayesian Optimization.

Area of Science:

• Biotechnology and Synthetic Biology
• Computational Biology and Bioinformatics

Background:

• Cell-free protein synthesis (CFPS) enables rapid protein production but faces challenges with yield variability due to volume constraints and batch inconsistencies.
• Existing optimization methods like Bayesian optimization (BO) are limited by fixed constraints and do not account for variability.

Purpose of the Study:

• To leverage Gaussian process (GP) predictive uncertainty for designing condition spaces in CFPS under variable constraints.
• To improve both exploitation (yield assurance) and exploration (Pareto front expansion) strategies in CFPS optimization.

Main Methods:

• Utilized GP predictive uncertainty, specifically lower and upper confidence bounds, for directed optimization in two CFPS systems.
• Developed a yield assurance space (YAS) for reliable condition selection under uncertainty.
• Implemented Pareto front expansion under tightened volume constraints using uncertainty-driven sampling.

Main Results:

• Achieved improved green fluorescent protein (GFP) yields in two CFPS systems within a few optimization rounds.
• Demonstrated the effectiveness of YAS in identifying conditions exceeding a conservative yield threshold.
• Showcased successful Pareto front expansion under altered volume constraints.

Conclusions:

• Gaussian process predictive uncertainty is a powerful tool for optimizing CFPS under variable constraints.
• This approach enables both assured condition design and targeted exploration for enhanced protein production.
• The methodology offers a robust framework for improving the consistency and efficiency of CFPS systems.