Divergent microbial signatures support accurate machine‑learning classification of hosts and parasite microbiomes

Abstract

Amphibians depend on their skin microbiome for essential functions including respiration, osmoregulation, and protection from pathogens. Ectoparasites such as soft ticks maintain close physical contact with their hosts, creating opportunities for microbial overlap and transfer. Accurately distinguishing these microbial communities is important for avoiding sample misclassification and for improving assessments of host health and pathogen dynamics. We hypothesized that distinct microbial signatures would allow machine learning (ML) models to classify host versus parasite samples using amplicon sequence variant (ASV) data. To test this, we analyzed microbiomes from frog skin and their associated ticks collected in Puerto Rico during 2012–2013. We implemented several ML algorithms to compare classification performance across models. All approaches showed high accuracy, with area under the curve (AUC) values above 0.9, and more complex models, such as random forests, frequently achieved perfect classification (AUC = 1.0). These results support our hypothesis that divergent microbiome signatures effectively differentiate hosts and parasites, reflecting biologically relevant community differences. Our framework has broader applications, including identifying microbial biomarkers of disease, classifying unknown environmental samples, and advancing studies of microbial transfer and co‑evolution in host–parasite systems.