Genetic algorithm-guided sample design enables efficient machine learning-driven optimization of tauroursodeoxycholic...

Area of Science:

• Biotechnology
• Machine Learning
• Metabolic Engineering

Background:

• Machine learning (ML) is powerful for optimizing fermentation but requires extensive data.
• Low-throughput microbial systems struggle to generate large datasets for ML.
• Existing experimental designs are often inefficient for complex bioprocesses.

Purpose of the Study:

• To present a sample-efficient, genetic algorithm (GA)-assisted adaptive sampling strategy for ML-driven fermentation optimization.
• To enable effective optimization of microbial biotransformation processes with limited experimental data.
• To demonstrate the utility of this strategy in a real-world biotransformation case study.

Main Methods:

• Employed a genetic algorithm (GA) for iterative selection of informative experimental conditions.
• Generated a compact dataset using GA-selected conditions for artificial neural network (ANN) modeling.
• Utilized an ANN-based response surface analysis to identify optimal fermentation parameters.
• Applied the GA-assisted adaptive sampling strategy coupled with ANN (GA-ANN) workflow for tauroursodeoxycholic acid (TUDCA) biotransformation in engineered Escherichia coli.

Main Results:

• The GA-ANN workflow successfully identified favorable fermentation medium compositions under a limited experimental budget.
• Achieved a maximum TUDCA/TCDCA ratio of 2.18 ± 0.06 and a TUDCA titer of 7.11 ± 0.12 g/L.
• ANN analysis revealed a biphasic effect of corn dextrin concentration on TUDCA conversion, highlighting nonlinear process behavior.

Conclusions:

• The GA-assisted adaptive sampling strategy offers a feasible and sample-efficient approach for low-throughput fermentation optimization.
• This method overcomes data limitations inherent in microbial fermentation systems.
• The study provides a practical workflow for optimizing bioprocesses with constrained experimental resources.