Batch growth of Kluyveromyces lactis cells from deproteinized whey: Response surface methodology versus Artificial neural network – Genetic algorithm approach

Deproteinized cheese making whey (CMW) was investigated as an alternative medium for the production of Kluyveromyces lactis as single-cell protein. Batch runs were performed according to a Full Factorial Design (FFD) on CMW supplemented with yeast extract, magnesium sulfate and ammonium sulfate in different concentrations. These independent variables were tested in duplicate at three levels, while dry biomass productivity was used as the response. The results were used to construct two models, one based on Response Surface Methodology (RSM) and another on Artificial Neural Network (ANN). Two different training methods (10-fold cross validation and training/testing) were utilized to obtain two different network architectures, while a genetic algorithm was utilized to obtain optimal concentrations of the above medium components. A quadratic regression by RSM (R2=0.840) was the best modeling and optimization tool under the specific conditions selected here. The highest biomass productivity (approximately 2.14 gDW/L h) was ensured by the following optimal levels: 7.04-9.99 % (w/v) yeast extract, 0.430-0.503 % (w/v) magnesium sulfate and 4.0 % (w/v) ammonium sulfate. These results demonstrate the feasibility of using CMW as an interesting alternative to produce single-cell protein.