Machine learning classification of quorum sensing-induced bacterial aggregation using flow rate assays on paper chips toward bacterial species identification in potable water sources

Preventing waterborne disease caused by bacteria is especially important in low-resource settings, where skilled personnel and laboratory equipment are scarce. This work reports a straightforward method for classifying bacterial species by monitoring the capillary flow rates on a multi-channel paper microfluidic chip, where quorum sensing (QS)-induced bacterial aggregation leads to measurable changes in flow rates, enabling species differentiation. It required no fluorescent molecules, microscope, particles, covalent conjugation, or surface immobilization. Five representative QS molecules and control were added to each bacterial sample, and their different extents of bacterial aggregation resulted in varied flow rates. Flow rates were collected for the duration of the flow to build the learning database, and the XGBoost machine learning algorithm predicted the accuracy for classifying ten bacterial species, including 7 gram-negative and 3 gram-positive species. Three different algorithms were developed for high, medium, and low bacterial concentration ranges, and the classification accuracies of all the algorithms exceeded 75.0 %. Using XGBoost and the previously established database, we tested bacteria in the field water samples and successfully predicted the dominant species. The technology developed in this study, using only QS molecules and a paper microfluidic chip, offers a simple system for detecting microorganisms in drinking water to help prevent waterborne diseases.