Comparison of predictive capabilities of response surface methodology and artificial neural network for optimization of periplasmic interferon-α2b production by recombinant Escherichia coli

Aim. Predictive capability of response surface methodology (RSM) and artificial neural network (ANN) in modeling growth of recombinant Escherichia coli (E. coli) and production of periplasmic interferon-alpha 2b (PrIFN)-alpha 2b was compared. Methods. The effect of potassium phosphate buffer at different pH and strengths, and EDTA concentration against the production of PrIFN-alpha 2b by E. coli were simulated using RSM and ANN. A central composite design (CCD) and multilayered feed-forward (MFF) incremental propagation algorithm of ANN were developed to predict the three variables. Results. In the validation results, the maximum cell growth (5.52 g/L) and production of PrIFN-alpha 2b (121.4 ng/mL) using RSM model were obtained at pH 7.13, 0.1 M phosphate buffer and EDTA concentration of 0.047 mM. On the other hand, the optimized condition derived by ANN model (dry cell weight of 5.43 g/L and PrIFN-alpha 2b production of 130.5 ng/mL) was obtained at pH 7.00, 0.124 M phosphate buffer and EDTA concentration of 0.07 mM. Conclusion. The results indicated that ANN had better accuracy and predictive capability as compared to RSM for the nonlinear behavior of data sets. The final optimized medium was 30% higher as compared to non-optimized phosphate buffer medium in shake flask culture.