Optimization of Process Parameters for the Production of Soy Protein by Ultrafiltration Using ANN

Abstract

The growing popularity of soy proteins among vegans and vegetarians, owing to their high protein content and widespread availability, has led to scientific studies on its various extraction methods mainly on ultrafiltration. This research employed artificial neural network (ANN) and Box-Behnken design (BBD) methodologies to predict the process parameters of ultrafiltration for the preparation of soy protein. Using BBD, the optimum process parameters of ultrafiltration were identified via the desirability function approach. The optimized permeate flux was 11.13 litres per hour (LPH) and 85.52% protein content in retentate. The identified ideal process parameters for ultrafiltration to achieve maximal protein retention encompassed a 10 kDa membrane module, a transmembrane pressure of 117 kPa (17 PSI), a volume concentration ratio of 3.5, diafiltration set at 1, and a flow rate of 65% of the pump capacity, exhibiting an absolute percent error value of 2.81. Employing these refined process parameters, the predicted value for protein retentate stood at 80.49%. The predictive accuracy of the model achieved an impressive 99.61% for protein retention. The ANN model effectively predicted the optimal ultrafiltration conditions, resulting in maximal protein retention and a protein content accuracy of 96.41% and 99.61%, respectively.