Predicting Food Sources of Listeria monocytogenes Based on Genomic Profiling Using Random Forest Model.

Area of Science:

• Foodborne illness surveillance
• Molecular epidemiology
• Machine learning applications in public health

Background:

• Listeria monocytogenes causes severe foodborne illness, with sporadic cases often lacking identified food sources.
• Current molecular surveillance relies on whole genome multilocus sequence typing (wgMLST) primarily for outbreak detection.

Purpose of the Study:

• To investigate the utility of machine learning models for attributing the food sources of Listeria monocytogenes isolates.
• To assess the performance of random forest models using wgMLST allele data for source attribution.

Main Methods:

• Compiled Listeria monocytogenes isolates with known food sources (dairy, fruit, meat, seafood, vegetable) from the PulseNet database.
• Applied random forest machine learning models to wgMLST allele data to predict isolate food sources.
• Evaluated model prediction accuracy across different food categories.

Main Results:

• The random forest model achieved an overall prediction accuracy of 49%, significantly outperforming naive prediction accuracy (28%).
• Prediction accuracy varied by food category, with the highest for meat (65%) and lowest for seafood (37%).

Conclusions:

• Machine learning, specifically random forest models, can effectively utilize high-resolution wgMLST data for Listeria monocytogenes source attribution.
• This approach offers a promising tool for identifying food sources of sporadic foodborne illnesses, enhancing public health interventions.