Hybrid experimental-machine learning framework for media optimization enhances antioxidant production in Micrococcus...

Culture medium composition is a key determinant of microbial antioxidant production, yet systematic optimization remains challenging due to nonlinear interactions among nutrients and the large experimental design space. In this study, a hybrid experimental-computational framework integrating orthogonal array design, neural-network surrogate modeling, and genetic algorithm optimization was developed for antioxidant production in Micrococcus endophyticus SS-1. A three-level L81 orthogonal array was used to evaluate ten nutritional and cultivation factors, generating a dataset for surrogate model training. The trained neural network was coupled with a genetic algorithm to explore the formulation space efficiently and to identify candidate media predicted to maximize antioxidant activity. Experimental validation confirmed that the genetic algorithm-optimized medium exhibited higher antioxidant activity than both the baseline medium and the best orthogonal-array formulation, as determined using the 2,2-diphenyl-1-picrylhydrazyl radical-scavenging assay. To account for the concentration-dependent nature of single-point scavenging measurements, dose-response experiments were performed and the half-maximal effective concentration values were determined. The experimentally validated genetic algorithm-optimized medium showed a reduced half-maximal effective concentration, indicating enhanced antioxidant potency rather than an assay-dependent increase in scavenging percentage.