Machine learning-based biological process optimization for low molecular weight welan gum production.
Journal:
International journal of biological macromolecules
PMID:
40101811
Abstract
This study focuses on optimizing the fermentation process for the production of low molecular weight welan gum (LMW-WG) using Sphingomonas sp. ATCC 31555 with glycerol as the sole carbon source. A series of single-factor experiments were conducted to identify six key influencing factors. The fermentation conditions were then modeled and optimized using a backpropagation artificial neural network (BP-ANN) in conjunction with particle swarm optimization (PSO). Under optimized conditions (glycerol 22.6 g/L, beef extract 4.3 g/L, KH₂PO₄ 3.8 g/L, MgSO₄ 0.3 g/L, CaSO₄ 0.5 g/L, and 10 % inoculation), a yield of 16.28 ± 2.58 g/L of LMW-WG was achieved. Kinetic modeling and metabolic pathway analysis further elucidated the critical roles of carbon and nitrogen metabolism in LMW-WG biosynthesis. The results indicate the superiority of the ANN-PSO approach in optimizing multivariable nonlinear systems, providing scientific evidence and technical support for the efficient production of LMW-WG, with significant potential for industrial applications.