Machine learning-optimized bioprocess for macroidin production by Lysinibacillus macroides and its biomedical applications.

Abstract

The quest for solutions to infectious diseases and life-debilitating disease states has been ongoing for centuries now. Natural products researches have revealed bioactive compounds of plant and microbial origin that offer solutions to health conditions but with poor yield. This study reports yield improvement of a novel macroidin bacteriocin through robust comparative process optimization involving statistical and machine learning approaches. Response surface methodology (RSM), artificial neural network (ANN), and extreme gradient boosting (XGBoost) models showed adequate fitting capabilities considering statistical indices and performance errors as: RSM (R² = 0.9389; MSE = 0.3877), ANN (R² = 0.9727; MSE = 0.1379) and XGBoost (R² = 0.8758; MSE = 0.6272). The ANN model, with superior prediction results, was further optimized by evolutionary (genetic algorithm-GA) and swarm (particle swarm optimization) intelligence techniques which increased macroidin concentration by 2.38-fold and 2.2-fold, respectively. ANN's superior parameter generalization and remarkable validation accuracy by GA at 23.1 °C, pH 8.89, 0.5 vvm aeration, and 248.6 rpm agitation selected the ANN-GA model for bioreactor production. The scale-up study revealed a volumetric oxygen transfer coefficient of 33.95 h^-1 at 250 rpm and 0.5 vvm, at which a macroidin yield, Y_p/x of 0.93 g g^-1 and productivity of 2.00 g L^-1 h^-1 were achieved. Evaluated pharmaco-clinical potentials of macroidin revealed significant (p < 0.05) anti-proliferative effects against HepG2 and MCF-7 cell lines and bactericidal and antibiofilm activities against ESKAPE pathogens. The bactericidal action was revealed to proceed through membrane permeability, electrolyte, and ATP depletion, to cell lysis.