Statistical and Artificial Neural Network Approaches to Modeling and Optimization of Fermentation Conditions for Production of a Surface/Bioactive Glyco-lipo-peptide

A freshwater alkaliphilic strain of Pseudomonas aeruginosa, grown on waste frying oil-basal medium, produced a surface-active metabolite identified as glycolipopeptide. Bioprocess conditions namely temperature, pH, agitation and duration were comparatively modeled using statistical and artificial neural network (ANN) methods to predict and optimize product yield using the matrix of a central composite rotatable design (CCRD). Response surface methodology (RSM) was the statistical approach while a feed-forward neural network, trained with Levenberg–Marquardt back-propagation algorithm, was the neural network method. Glycolipopeptide model was predicted by a significant (P < 0.001, R(2) of 0.9923) quadratic function of the RSM with a mean squared error (MSE) of 3.6661. The neural network model, on the other hand, returned an R(2) value of 0.9964 with an MSE of 1.7844. From all error metrics considered, ANN glycolipopeptide model significantly (P < 0.01) outperformed RSM counterpart in predictive modeling capability. Optimization of factor levels for maximum glycolipopeptide concentration produced bioprocess conditions of 32 °C for temperature, 7.6 for pH, agitation speed of 130 rpm and a fermentation time of 66 h, at a combined desirability function of 0.872. The glycosylated lipid-tailed peptide demonstrated significant anti-bacterial activity (MIC = 8.125 µg/mL) against Proteus vulgaris, dose-dependent anti-biofilm activities against Escherichia coli (83%) and Candida dubliniensis (90%) in 24 h and an equally dose-dependent cytotoxic activity against human breast (MCF-7: IC50 = 65.12 µg/mL) and cervical (HeLa: IC50 = 16.44 µg/mL) cancer cell lines. The glycolipopeptide compound is recommended for further studies and trials for application in human cancer therapy.