Application of Artificial Neural Networks to Predict Inhibition in Probiotic Experiments

Abstract

Artificial neural networks (ANNs) provide a modeling approach that can be used in the in vitro stages of probiotic studies. The aim of the study was to evaluate the ability of multilayer perceptron (MLP) and radial-basis function (RBF) ANNs to predict the inhibition level of indicator bacteria in co-culture experiments performed at various initial concentrations. In both types of networks, time, initial concentrations of L. lactis and Aeromonas spp. were the input variables and the inhibition concentration of Aeromonas spp. was the output value. In the construction of the models, different numbers of neurons in the hidden layer, and different activation functions were examined. The performance of the developed MLP and RBF models was tested with root mean square error (RMSE), coefficient of determination (R 2 ) and relative error (e) statistical analysis. Both ANN models were showed a strong agreement between the predicted and experimental values. However, the developed MLP models showed higher accuracy and efficiency than the RBF models. The results indicated that ANNs developed in this study can successfully predict the inhibition concentration of Aeromonas spp. co-cultured with L. lactis in vitro and can be used to determine bacterial concentrations in the design of further experiments.