Multi-dimensional Optimization of High-density Lactobacillus gasseri GU-G23 Culture

LIAO Yanting,LI Ruiding,HUANG Guohong,LIANG Xiaolin,ZHOU Fan,MA Zhize,SHI Fengcui,LI Quanyang

DOI: https://doi.org/10.13982/j.mfst.1673-9078.2023.10.1326

2023-01-01

Abstract:In order to improve the biomass of Lactobacillus gasseri GU-G23 with probiotic properties and obtain its high-density culture factors, this paper firstly screened the main growth nutrient factors of this strain as fish peptone, tryptone and triammonium citrate by single-factor experiment and Plackett-Burman experiment. The response surface experimental design combination was used as the training sample for the machine model, and the random forest regression（RFR） and radial basis function neural network（RBF） models were used to predict their culture media formulations. Using the coefficient of determination（R-squared, R~2）, Mean absolute deviation（MAE）, Mean-square error（MSE） and Mean absolute percentage error（MAPE） as model evaluation metrics, the comparison concluded that RBF had better predictive performance in this study. Subsequently, a combination of RBF neural network and Genetic algorithm（GA） was chosen to optimize the main components of the culture medium. The optimized formulation of the medium was obtained as follows: fish peptone 29.89 g/L, tryptone 23.33 g/L, triammonium citrate 4.34 g/L, sucrose 15.00 g/L, xylo-oligosaccharides 15.00 g/L, sodium acetate 5.00 g/L dipotassium hydrogen phosphate 0.40 g/L, magnesium sulfate heptahydrate 0.58 g/L, manganese sulfate monohydrate 0.29 g/L, Tween-80 1.00 g/L. Under this medium condition, the number of viable bacteria in the resulting sample reached 5.21×10~9 CFU/mL, which was 4.57 times higher than that before optimization. This study provides a new approach to the optimization of microbial high-density culture prediction.