Machine Learning for Improved Detection of Pathogenic <i>E. coli</i> in Hydroponic Irrigation Water Using Impedimetric Aptasensors: A Comparative Study

Reuse of alternative water sources for irrigation (e.g., untreated surface water) is a sustainable approach that has the potential to reduce water gaps, while increasing food production. However, when growing fresh produce, this practice increases the risk of bacterial contamination. Thus, rapid and accurate identification of pathogenic organisms such as Shiga-toxin producing <i>Escherichia coli</i> (STEC) is crucial for resource management when using alternative water(s). Although many biosensors exist for monitoring pathogens in food systems, there is an urgent need for data analysis methodologies that can be applied to accurately predict bacteria concentrations in complex matrices such as untreated surface water. In this work, we applied an impedimetric electrochemical aptasensor based on gold interdigitated electrodes for measuring <i>E. coli</i><i>O157:H7</i> in surface water for hydroponic lettuce irrigation. We developed a statistical machine-learning (SML) framework for assessing different existing SML methods to predict the <i>E. coli</i><i>O157:H7</i> concentration. In this study, three classes of statistical models were evaluated for optimizing prediction accuracy. The SML framework developed here facilitates selection of the most appropriate analytical approach for a given application. In the case of <i>E. coli</i><i>O157:H7</i> prediction in untreated surface water, selection of the optimum SML technique led to a reduction of test set RMSE by at least 20% when compared with the classic analytical technique. The statistical framework and code (open source) include a portfolio of SML models, an approach which can be used by other researchers using electrochemical biosensors to measure pathogens in hydroponic irrigation water for rapid decision support.