A combinatorial optimization platform was used to enhance soluble production of Pseudomonas aeruginosa exotoxin A (ETA) in engineered Escherichia coli. Twelve strains were screened across four induction temperatures, three chaperone systems, and four redox-modulating additives, totaling 576 conditions.
A high-throughput fluorescence-based solubility reporter enabled rapid measurement of soluble ETA yields, which were then modeled with an XGBoost machine learning algorithm. Model training employed 5-fold cross-validation with hyperparameter optimization to reduce overfitting.
Statistical analyses included one-way ANOVA with Tukey post-hoc testing, Pearson correlation, and multiple regression. The study identified a disulfide-competent SHuffle T7 strain induced at 12°C, co-expressing DnaKJE/GroEL chaperones and supplemented with 2 mM oxidized glutathione, as yielding 3.24 ± 0.4 mg/L soluble, enzymatically active ETA.
This corresponds to a 15-fold increase relative to conventional BL21(DE3) systems (statistical F value reported as 45.32 in the text). The authors present an integrated experimental–ML platform that combines high-throughput screening with predictive modeling to address soluble production challenges of disulfide-rich proteins, noting that generalizability to other targets requires further validation.
coli.
coli strains across a matrix of four induction temperatures, three chaperone systems, and four redox-modulating additives.