Neural networks and genetic algorithm as robust optimization tools for modeling the microbial production of poly-ß-hydroxybutyrate (PHB) from Brewers' spent grain.

This research work has been carried out to establish the combinatorial impact of various fermentation medium constituents, used for poly-ß hydroxybutyrate (PHB) biosynthesis. Model development was performed with an optimized medium composition that enhanced the biosynthesis of PHB from the biowaste material Brewers' spent grain (BSG). The latter was used as a carbon substrate in submerged fermentation with Bacillus sphaericus NCIM 2478. Three independent variables: BSG, yeast extract (YE), and salt solution concentration (SS) and one dependent variable (amount of PHB produced) were assigned. A total of 35 microbial fermentation trials were conducted by which a nonlinear mathematical relationship was established in terms of neural network model between independent and dependent variables. The resulting artificial neural networks (ANNs) model for this process was further optimized using a global genetic algorithm optimization technique, which predicted the maximum production of PHB (916.31 mg/L) at a concentration of BSG (50.12 g/L), concentration of YE (0.22 g/L), and concentration of SS (24.06%, v/v). The experimental value of the quantity of PHB (concentration ∼916 mg/L) was found to be very close to the value predicted by the ANN-GA model approach.