LAUSR

Abstract Inhibitors are an important means of reducing the growth of foodborne microorganisms. However, excessively using inhibitors will cause great harm to human beings and organisms. Therefore, quantitatively evaluating the effect of low-dose inhibitors on foodborne bacteria growth is very important. Electrochemical impedance spectroscopy (EIS) is a commonly used biosensor technique for detecting foodborne bacteria. In this paper, we show a machine learning-based EIS biosensor method that can be used to detect the effect of a low-dose inhibitor (e.g., hydrogen peroxide) on Escherichia coli (E. coli). After obtaining the minimum inhibitory concentration (MIC), the inhibitor concentration and below was applied to E. coli solution. EIS data were obtained by binding the target bacteria to the electrode surface through antibodies, and then, the impedance parameters were fitted by the Randles model. XGBoost and a support vector regression (SVR) machine learning models were compared to establish a quantitative relationship between multiple impedance parameters and the bacterial concentration under the effect of inhibitors. The results showed that XGBoost improved the quantitative analysis accuracy toward inhibitors (0.175–0.375 μL/mL), based on the bacteria growth in the solution. After different low concentrations of the inhibitor were added into a standard bacterial solution for 1 h, 2 h and 3 h, the maximum prediction error of the inhibitor concentration was 4.95%, 1.03% and 0.46%, respectively. The prediction error decreased as the incubation time of the E. coli culture was extended. These results pave the way for the automation of an accurate EIS biosensor for analyzing foodborne microorganisms under various low doses of inhibitors or drugs