Leveraging machine learning to uncover multi-pathogen infection dynamics across co-distributed frog families

Area of Science:

• Conservation Biology
• Wildlife Disease Ecology
• Amphibian Pathology

Background:

• Amphibian populations face significant declines due to emerging infectious diseases.
• Understanding pathogen dynamics is complex, influenced by host, pathogen, and environmental factors.
• Traditional statistical methods struggle with multi-variable analyses across broad scales.

Purpose of the Study:

• To leverage natural history collections and machine learning to analyze amphibian pathogen dynamics.
• To identify key drivers of infection prevalence and intensity for three generalist frog pathogens.
• To assess the influence of host traits and environmental variables on infection patterns.

Main Methods:

• Sampled 12 focal amphibian taxa across the eastern USA from frozen natural history collections.
• Quantified infection loads using qPCR for Batrachochytrium dendrobatidis (Bd), Ranavirus (Rv), and Amphibian Perkinsea (Pr).
• Applied balanced random forests (RF) models to predict infection status and intensity based on host and environmental data.

Main Results:

• Approximately 20% of sampled amphibians were infected; Bd was most prevalent (16.9%), followed by Rv (4.38%) and Pr (1.06%).
• Ranidae family showed highest prevalence and intensity, particularly for Rv and Bd.
• Host species and cooler, stable temperatures at higher latitudes were key predictors for Bd and Rv; Pr trends were opposite. Juveniles had higher Rv and Bd loads than adults.

Conclusions:

• Machine learning and broad sampling effectively identify drivers of amphibian infections.
• Host taxonomy and temperature are critical factors influencing pathogen dynamics in amphibians.
• Findings inform conservation strategies by highlighting key ecological correlates of disease risk.