Summary
A Salmonella-positive result in agricultural water may not always carry the same level of concern. Some Salmonella types appear repeatedly in rivers and ponds without clear links to illness, while others are commonly linked to human illness. This one-year pilot project will test whether large sequence datasets, combined with environmental context (e.g., season, location, and rainfall), can identify genetic markers that distinguish lower-concern background populations from those linked to human illness, better than subtype labels alone. We will combine Salmonella genomes from an Eastern U.S. watershed study, public databases that link related water and clinical strains, and a long-term California monitoring program, then use advanced data analytics as a practical “pattern-finder” that learns the underlying relationships between genetic signatures, environment, and clinical linkage. We will screen historical California isolates using rapid methods and confirm a subset with targeted sequencing to refine a short list of predictive targets, then apply the refined list to newly collected seasonal field samples in California. The expected benefit is a cost-efficient way to interpret Salmonella detections when testing occurs, helping focus follow-up actions and confirmatory testing where they are most likely to matter, and providing a foundation for future expansion as more datasets become available.
Technical Abstract
Surface waters used for agricultural production can contain diverse Salmonella enterica populations, yet serovar labels and a binary Salmonella-positive signal may be insufficient to consistently distinguish isolates that overlap with clinically linked lineages from recurring background environmental populations. The objective of this one-year pilot project is to identify Salmonella sequence targets that support context-aware risk profiling of agricultural water detections using artificial intelligence (AI) analytics. The project is organized into three integrated tasks: (1) assemble a harmonized, model-ready genomic and contextual dataset by integrating an established Mid-Atlantic surface water isolate collection with environmental metadata, curated public genomes from the NCBI (National Center for Biotechnology Information) Pathogen Detection resource linking water and clinical SNP (single-nucleotide polymorphism) clusters, and a long-term California surface water monitoring dataset; (2) establish a region-adaptable analytics framework that combines unsupervised analyses (to summarize genomic structure and screen for technical artifacts) with supervised models trained on genomic and contextual inputs to identify and rank a short list of candidate sequence targets. This framework will also quantify how target stability and model performance change as additional California genomes are incorporated, including practical assessment of data sufficiency (i.e., how much data is enough for reliable results); and (3) validate targets using practical laboratory methods and independent datasets by refining a qPCR (quantitative polymerase chain reaction) panel through scalable screening of historical California isolates with confirmatory targeted amplicon sequencing at MSU, then applying the refined panel to newly collected seasonal field isolates to test whether model-identified higher-risk contexts correspond to increased detection of predictive targets. Model performance will be compared against serovar-only baselines using precision, recall, and area under the precision-recall curve, with an anticipated improvement on the order of 20% in precision-recall performance on independent California datasets. Overall, the project will deliver a harmonized dataset (≥1,000 isolates), a ranked list of 10–15 testable sequence targets, and a validated framework linking genomic signals and environmental context to more informative interpretation of detection events. These outcomes will enable more targeted confirmatory testing and establish a foundation for partners to extend the approach as additional datasets become available.
Research Objectives
The objective of this one-year pilot project is to identify Salmonella sequence targets that support context-aware risk profiling of agricultural water detections using AI analytics. To achieve this objective, the project is organized into three tasks (Fig. 1): 1) assemble a harmonized, multi-source dataset and candidate sequence target library; 2) establish a region-adaptable AI framework for predictive sequence target discovery, and 3) validate predictive targets using historical and new datasets.
Findings & Recommendations
This project is ongoing. A final report will be provided when the project is finished.