Statistical versus neural network-embedded swarm intelligence optimization of a metallo-neutral-protease production: activity kinetics and food industry applications.

Abstract

An integrated approach involving response surface methodology (RSM) and artificial neural network-ant-colony hybrid optimization (ANN-ACO) was adopted to develop a bioprocess medium to increase the yield of Bacillus cereus neutral protease under submerged fermentation conditions. The ANN-ACO model was comparatively superior (predicted r² = 98.5%, mean squared error [MSE] = 0.0353) to RSM model (predicted r² = 86.4%, MSE = 23.85) in predictive capability arising from its low performance error. The hybrid model recommended a medium containing (gL^-1) molasses 45.00, urea 9.81, casein 25.45, Ca²⁺ 1.23, Zn²⁺ 0.021, Mn²⁺ 0.020, and 4.45% (vv^-1) inoculum, for a 6.75-fold increase in protease activity from a baseline of 76.63 UmL^-1. Yield was further increased in a 5-L bioreactor to a final volumetric productivity of 3.472 mg(Lh)^-1. The 10.0-fold purified 46.6-kDa-enzyme had maximum activity at pH 6.5, 45-55 °C, with K_m of 6.92 mM, V_max of 769.23 µmolmL^-1 min^-1, k_cat of 28.49 s^-1, and k_cat/K_m of 4.117 × 103 M^-1 s^-1, at 45 °C, pH 6.5. The enzyme was stabilized by Ca²⁺, activated by Zn²⁺ but inhibited by EDTA suggesting that it was a metallo-protease. The biomolecule significantly clarified orange and pineapple juices indicating its food industry application.