Application of artificial neural network coupling particle swarm optimization algorithm to biocatalytic production of GABA

The biotransformation of L-sodium glutamate (L-MSG) to gamma-aminobutyric acid (GABA) catalyzed by the cells of Lactobacillus brevis with higher glutamate decarboxylase activity was investigated. The results showed that pH, temperature, and FeSO4 (.) 7H(2)O concentration had significantly positive effect on GABA yield. The individual and interactive effects of pH, temperature, and FeSO4 (.) 7H(2)O concentration were further optimized in terms of GABA yield. In the present work; an artificial neural network (ANN) and response surface methodology (RSM) models were developed, which incorporated pH, temperature, and FeSO4 (.) 7H(2)O concentration as input variables, and GAGA yield as output variable. The optimized ANN topology included four neurons in the hidden layer and the best network architecture was 3-4-1. The trained ANN gave total root-mean square error (sigma) equal to 1.84 for GABA yield while the RSM gave a equal to 2.63. The results demonstrated a slightly higher prediction accuracy of ANN compared to RSM. The modeled maximum GABA yield was identified by applying particle swarm optimization algorithm to the ANN model developed. The modeled maximum GABA yield reached 91 mM under the following optimal conditions: 25 ml, Na2HPO4-citric acid buffer (100 mM, pH 4.23), 120 MM L-MSG, 0.83 g/L FeSO4 (.) 7H(2)O, 10 mu M PLP, the resting cells obtained from a 60-h culture broth, 2.68 g dry cell weight (DCW)/L, and without agitation at 40 degrees C for 5 h. The previous high value of GABA yield that was observed was 81.8 mM. The optimized conditions allowed GABA yield to be increased from 81.8 to 90.57 mM after verification experiments test.