Computational Surveillance of Microbial Water Quality With Online Flow Cytometry

Abstract

Automated flow cytometry (FCM) adapted to real-time quality surveillance provides high-temporal-resolution data about the microbial communities in a water system. The cell concentration calculated from FCM measurements indicates sudden increases in the number of bacteria, but can fluctuate significantly due to man-made and natural dynamics; it can thus obscure the presence of microbial anomalies. Cytometric fingerprinting tools enable a detailed analysis of the aquatic microbial communities, and could distinguish between normal and abnormal community changes. However, the vast majority of current cytometric fingerprinting tools use offline statistical computations which cannot detect anomalies immediately. Here, we present a computational model, entitled Microbial Community Change Detection (MCCD), which transforms microbial community characteristics into an online process control signal (herein called outlier score) that remains close to zero if the microbial community remains stable and increases with fluctuations in the community. The model is based on fingerprints and distance-based outlier calculations. We tested it in silico and in vitro by simulating acute contaminations to real-world water systems with large inherent microbial fluctuations. We showed that the outlier score was robust against these dynamic variations, while reliably detecting intentional contaminations. This model can be used with automated FCM to quickly detect potential microbiological contamination, and this especially when the time between treatment and distribution is very short.