New approach for modeling generalized microbial growth curves using artificial neural networks

Microbial growth curves are essential components in microbiological studies and are modeled conventionally by nonlinear fitting to one analytical expression such as the modified Gompertz equation. This paper discusses the potential of artificial neural networks (ANNs) for modeling bacterial growth curves. These ANNs are efficient approximators for highly dimensional complex functions because of their high nonlinearity and tolerance to noisy data. Therefore, ANNs can provide great flexibility in developing generalized models by extracting the real behavior directly from the experimental data. Such models can be designed to include the effect of time as well as a multitude of parameters pertaining to experimental conditions. The approach was applied to modeling time-dependent growth curves of Escherichia coli O157:H7 as affected by sodium chloride concentration and of Shigella flexneri as affected by incubation temperature, pH, and initial count. The developed ANNs were able to approximate the experimental growth curves with high accuracy. The advantages as well as limitations of the proposed methodology are presented.