Confronting data sparsity to identify potential sources of Zika virus spillover infection among primates

Area of Science:

• Epidemiology
• Machine Learning
• Zoonotic Disease Surveillance

Background:

• The Zika virus (ZIKV) epidemic highlighted the threat of mosquito-borne flaviviruses.
• Sylvatic cycles in primates pose spillover risks to humans, but data on reservoirs is scarce.
• Limited surveillance data complicates identifying and mitigating zoonotic disease threats.

Purpose of the Study:

• To address data sparsity in identifying zoonotic reservoirs.
• To develop computational methods for predicting high-risk animal species.
• To improve epidemiological response and prevention strategies for emerging zoonoses.

Main Methods:

• Employed Bayesian multi-label learning, a machine learning technique.
• Utilized multiple imputation to handle missing primate trait data.
• Developed predictive models for flavivirus-positive primates.

Main Results:

• Models achieved 82% accuracy in distinguishing flavivirus-positive primates.
• Identified primate species with high spillover risk are often adapted to human environments.
• Demonstrated the utility of computational methods in data-scarce zoonotic scenarios.

Conclusions:

• Machine learning effectively extracts actionable insights from limited data.
• Predictive modeling can guide surveillance and prevention of zoonotic diseases.
• Understanding host adaptation is key to mitigating spillover risk.