Response surface methodology and artificial neural network based media optimization for pullulan production in Aureobasidium pullulans

The selection and optimization of carbon and nitrogen sources are essential for enhancing pullulan production in Aureobasidium pullulans. In this study, combinations of carbon (sucrose, fructose, glucose) and nitrogen sources ((NH₄)₂SO₄, urea, NaNO₃) were screened, where sucrose and NaNO₃ offered the highest pullulan yield (9.33 g L⁻¹). Plackett–Burman design of experiment identified KH₂PO₄, NaCl, and sucrose as significant factors, which were further optimized using a central composite design. A hyperparameter-optimized artificial neural network (ANN) model with a 3-6-2-1 architecture demonstrated superior predictive accuracy (R²: 0.96) and generalizability (R²_CV: 0.74) over a reduced quadratic model (R²: 0.82). The predicted pullulan yield (31.9 g L⁻¹) under ANN model optimized conditions (sucrose: 79.9 g L⁻¹, KH₂PO₄: 0.25 g L⁻¹, NaCl: 4.3 g L⁻¹) closely matched with the observed yield (30.17 g L⁻¹), while quadratic model showed a significant deviation (39.7 g L⁻¹ vs. 21.0 g L⁻¹), highlighting the reliability of the ANN model.