Machine learning–driven discovery of host genetic factors for paratuberculosis in goats within the one health framework

Paratuberculosis, caused by Mycobacterium avium subsp. paratuberculosis (MAP), remains a persistent One Health concern due to its slow clinical course, environmental resilience, wide circulation in ruminant systems, and unresolved zoonotic implications. To characterise MAP exposure across Türkiye’s goat populations, we conducted a nationwide genomic survey encompassing seven breeds from 36 farms in 11 provinces. High-density SNP genotyping combined with mutual-information–based feature preselection retained informative, non-redundant loci capturing both linear and nonlinear components of disease architecture. Nine complementary machine-learning models were applied to identify host genetic factors underlying MAP infection, and an ensemble importance framework resolved 31 FDR-controlled SNPs consistently associated with MAP status. Functional annotation implicated immune processes including cytokine–receptor signalling, antigen presentation, glycan-mediated T-cell regulation, and NF-κB-linked inflammation. Concordance with mixed linear models and genome-wide McNemar tests suggested that both additive and non-additive genetic effects shape the observed signal. These reproducible, albeit preliminary, markers outline a genomic foundation for breeding MAP-resilient goats and point to opportunities for reducing pathogen shedding at its source within a broader One Health strategy.