Artificial Neural Network Modeling to Predict the Non-Linearity in Reaction Conditions of Cholesterol Oxidase from <i>Streptomyces olivaceus</i> <i>MTCC</i> 6820

Abstract

Cholesterol oxidase (COX) is widely used enzyme for total cholesterol estimation in human serum and for the fabrication of electro-chemical biosensors. COX is also used for the bioconversion of cholesterol; for the production of precursors of steroidal drugs and hormones. Enzyme activity depends decisively on defined conditions with respect to pH, temperature, ionic strength of the buffer, substrate concentration, enzyme concentration, reaction time. Standardization of these parameters is desirable to attain optimum activity of the enzyme. The present work aims to build a neural network model using five input parameters (pH, cholesterol concentration, 4-aminoantipyrine concentration, crude COX volume and horseradish peroxidase) and one output i.e., COX activity (U/ml) as a response. A feed forward back propagation neural network with Levenberg-Marquardt training algorithm was used to train the network. The network performance was assessed in terms of regression (R2), Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE). A network topology of 5-10-1 was found to be optimum. The MSE, MAPE and R2 values of the neural model were 0.0075%, 0.12% and 0.9792% respectively. The maximum predicted activity of COX was 1.073 U/ml, which was close to the experimental value i.e., 1.1 U/ml at simulated optimum assay conditions. MSE and MAPE depicted the precision in the prediction efficiency of the developed ANN model. Higher R2 value showed a good correlation between the experimental and ANN predicted values. This proved the robustness of the ANN model to predict similar type of system (COX from other Streptomyces sp.) within the limits of the trained data set. The COX activity was enhanced by 1.71 folds after optimization of the reaction conditions.