Abstract
This study systematically optimized the fermentation process for fructosyltransferase (FTase) production by Aspergillus niger FS054, integrating traditional experimental designs with machine learning approaches. Single-factor experiments initially identified critical medium components (carbon source, nitrogen sources, phosphate, and metal ions) and cultivation parameters (pH, liquid volume, inoculum size, temperature, and shaking speed). Subsequent Plackett-Burman screening identified sucrose, yeast extract paste, and NH4Cl as the most influential medium factors. Through Box-Behnken response surface methodology (RSM), the optimal medium composition was determined as sucrose 156.65 g/L, yeast extract paste 42 g/L, and NH4Cl 1.68 g/L, yielding an enzyme activity of 3249.00 ± 24.39 U/L (99.16% agreement with RSM predictions). Further optimization of cultivation conditions using a hybrid backpropagation neural network-genetic algorithm (BP-GA) model identified optimal parameters as pH 5.5, a liquid volume of 96.6 mL (in a 250 mL shaker), and inoculum size of 2.4 × 104 spores/mL, achieving a final enzyme activity of 3422.14 ± 36.86 U/L (1.1% deviation from the predicted 3460 U/L), representing a 4.2-fold increase over initial conditions. This work demonstrates the synergistic application of classical experimental design and artificial intelligence, significantly enhancing FTase productivity and potentially offering a more economical enzyme source for industrial-scale fructooligosaccharide (FOS) biosynthesis.