Metabolism-Based Bacterial Detection in Water Using First-Principles Modeling and Predictive Machine Learning

Bacterial contamination in water remains a global health concern. This study introduces a simplified and scalable approach for detecting Escherichia coli by coupling bacterial metabolism with predictive modeling. We developed a first-principle model to examine how E. coli consumes glucose as growth and metabolic rates vary over time. Detection time was primarily governed by bacterial growth rates, which exerted a substantially greater influence than variations in glucose consumption. Experimental validation on laboratory and environmental samples was performed using repurposed commercial glucose meters and portable pH sensors to monitor glucose depletion and proton production as indicators of bacterial viability. The model accurately captured these changes and fitted experimental data across varying growth rates induced by temperature differences (R2 ≥ 0.96, RMSE < 0.16 mM). Under optimal conditions, detection occurred within 3 h compared to 23 h under unfavorable conditions. Mixed cultures of Escherichia coli and Bacillus subtilis showed only a marginal effect on detection time relative to temperature. Selective detection of Escherichia coli was achieved using nutrient-specific media. Integrating metabolic data with pH, temperature, and time into machine-learning models enhanced predictive capability, with the decision tree model achieving a mean squared error of 0.0802 and identifying pH, time, and temperature as dominant predictors. Overall, this work redefines bacterial culture as a data-driven, field-ready diagnostic system, transforming simple consumer devices into quantitative biosensors for rapid, low-cost microbial monitoring in environmental and public-health applications.