The joint effect of ultrasound and magnetic Fe3O4 nanoparticles on the yield of 2,6-dimethoxy-rho-benzoquinone from fermented wheat germ: Comparison of evolutionary algorithms and interactive analysis of paired-factors

2,6-Dimethoxy-p-benzoquinone (DMBQ) is a potential anti-tumor substance found in the fermented wheat germ. In this study, ultrasound and Fe3O4 nanoparticles were used to improve the DMBQ yield. An artificial neural network (ANN) embedded separately with the back-propagation algorithm (BP), genetic algorithm (GA), particle swarm optimized algorithm (PSO), ant colony optimized algorithm (ACO), GA-ACO, GA-PSO and PSO-ACO, were used to establish the relationship between 11 factors and DMBQ yield. The robustness and generalization of PSO-ACO-ANN, which gave the minimum mean squared error and mean absolute percentage error for the training and WA dataset, was superior to the others. Next, a modified Garson's algorithm and mixed partial derivatives algorithm indicated that the most influential paired-parameters were ultrasonic power and concentration of nanoparticles. Finally, the factors were optimized by six optimization algorithms, and confirmatory experimental results indicated that the optimum DMBQ yield was 0.213 +/- 0.007 mg/g, which was 161.2% higher than the control.