Conditional Forest Models Built Using Metagenomic Data Accurately Predicted Salmonella Contamination in Northeastern...

Area of Science:

• Environmental microbiology
• Food safety
• Bioinformatics

Background:

• Contaminated water used in agriculture poses a significant risk of foodborne illnesses due to pathogens like Salmonella.
• Current methods for detecting agricultural water contamination are insufficient, necessitating novel approaches for pathogen assessment.

Purpose of the Study:

• To evaluate the effectiveness of water microbiome data in predicting Salmonella contamination in streams used for produce irrigation.
• To identify potential microbial indicators of Salmonella contamination in agricultural water sources.

Main Methods:

• Collected water samples from 60 New York streams and tested for Salmonella presence.
• Performed Illumina shotgun metagenomic sequencing to analyze the water microbiome composition.
• Utilized machine learning models (Conditional Forest, Regularized Random Forest, Support Vector Machine) to predict Salmonella contamination based on microbiome data.

Main Results:

• Conditional Forest models demonstrated high accuracy (AUC 0.86, Kappa 0.53) in predicting Salmonella contamination.
• Aeromonas salmonicida and Aeromonas sp. strain CA23 were identified as the most significant taxa for predicting Salmonella presence.
• These findings align with differential abundance tests, highlighting Aeromonas species as key indicators.

Conclusions:

• Water microbiome data, analyzed with machine learning, can effectively predict Salmonella contamination in agricultural water.
• Aeromonas species show promise as novel indicators for Salmonella, potentially improving water safety monitoring strategies.
• Further research is needed to validate these indicators across diverse environmental conditions and develop rapid detection assays.