Ichnaea-MST: An Automated Machine Learning Tool for Microbial Source Tracking providing prediction of Faecal Source of Contamination in Water

Faecal contamination in water is a persistent global challenge with significant public health and environmental implications. Microbial Source Tracking (MST) seeks to determine whether pollution originates from human or animal sources. However, conventional approaches relying on individual markers and/or microbial indicators and expert interpretation often fall short in complex scenarios, due mainly to low abundance of targets, diluted water samples or environmentally degraded conditions. An open-source software platform, Ichnaea-MST, that applies machine learning (ML) to MST datasets to enhance source attribution accuracy has been developed. By integrating the H2O AutoML framework, Ichnaea-MST automates data augmentation, model training, validation, and classification, thus enabling microbiologists to exploit advanced ML algorithms without requiring programming expertise. Ichnaea-MST accommodates diverse MST-markers, including bacterial, viral, mitochondrial, and chemical indicators, and accounts for environmental persistence through T 90 decay modelling and generates probabilistic classifications with associated performance metrics. Demonstrative datasets from European research projects were used to assess both binary (Human vs. Non-Human) and multinomial (Human, Bovine, Porcine, Poultry) classification scenarios. In this evaluation, Ichnaea-MST was trained and validated using single-label (i.e., one source assigned per sample) reference samples; therefore, the reported performance metrics refer to supervised single-label source classification rather than quantitative mixed-source apportionment. Ichnaea-MST demonstrated high predictive accuracy (0.907) in a 5-class multinomial classification task, with macro-averaged and micro-averaged area under the receiver operating characteristic (ROC) curve (AUC) of 0.989 and 0.994, respectively. Application examples illustrate its utility for drinking-water safety assessments, probabilistic source classification, marker-panel evaluation, and scenario-based interpretation of environmentally aged samples.