Accessory-based source attribution (AB_SA)

Accessory-based source attribution (AB_SA) attributes the animal or environmental sources of bacterial isolates using accessory-genome features and a multinomial logistic regression classifier for epidemiological source attribution.

Key Features:

• Multinomial Logistic Regression Classifier: AB_SA employs a multinomial logistic regression classifier to predict the animal source of bacterial isolates using genomic data.
• Source-Enriched Loci Extraction: The method utilizes source-enriched loci extracted from accessory-genome profiles within a pangenomic dataset to inform classification.
• Model Training and Self-Attribution: The model is trained on strains with known animal-source categories (e.g., poultry, ruminant, pig) and incorporates a self-attribution step to select the optimal number of candidate accessory genes.
• Predictor Gene Identification: In an application to Salmonella enterica Typhimurium and its monophasic variant (S. enterica 1,4,[5],12:i:-), AB_SA identified eight predictor genes among 2802 accessory genes.
• Predictive Performance: The method demonstrated 80% self-attribution accuracy and classified 25 of 29 isolates from unknown sources with over 85% probability in the reported application.

Scientific Applications:

• Epidemiological Source Attribution: Assigning human or environmental isolates to animal sources to support outbreak investigation and surveillance.
• Tracing Contamination Pathways: Tracing likely animal-origin contamination within food production chains.
• Environmental Reservoir Identification: Identifying environmental reservoirs of pathogenic bacteria based on accessory-genome signatures.

Methodology:

Extracting accessory-genome profiles from a pangenomic dataset; selecting source-enriched loci; training a multinomial logistic regression classifier on known-source strains; performing self-attribution to determine the optimal number of accessory genes; and identifying key predictor genes for source classification.