Artificial neural networks as a tool for incorporating microbial stress adaptations in the quantification of microbial inactivation

Abstract

Quantifying microbial adapted responses due to thermal stresses by an accurate methodology is imperative for assessing the efficacy of a heat process. Two different artificial neural network (ANN) models are constructed for studying the increased induction of the heat resistance of Escherichia coli K12 under a treatment of decreasing heating rates. In the first model structure there are two input vectors, namely, time t and temperature rate dT=dt, whereas in the second case is also added a third one, namely, the microbial load delayed with one time unit Nk¡1. For both models a minimal fully-connected feedforward architecture is used consisting of one hidden neuron and one output neuron. Results as based on the prediction capability of the model structures demonstrate the comparative advantage when an ANN architecture with a delay in its inputs is employed. Incorporation of past events seems to be an essential input for taking into account the observed induced microbial heat resistance.