HantaWatch uses federated learning for hantavirus genomic surveillance

A new paper on arXiv (2607.16234) introduces HantaWatch, a federated learning framework designed to overcome key barriers in hantavirus genomic surveillance. Current approaches are limited by the distribution of sequence data, non-IID (non-identically distributed) source heterogeneity, and constrained expert-review capacity. HantaWatch addresses these by enabling laboratories and surveillance sites to collaboratively train sequence-based models without ever sharing raw genomic data. The framework integrates several technical components: k-mer feature extraction, source-aware federated client construction, adaptive DU-FedProx optimization, surveillance-specific model selection, and prediction-only triage.

Experiments on both binary and multi-class tasks demonstrate that HantaWatch can support high-risk screening, outbreak-associated prediction, clade classification, and clinical-syndrome categorization. Critically, it balances predictive performance, false-negative risk, and update stability. The model’s outputs are converted into actionable metrics: risk scores, confidence estimates, uncertainty flags, and ranked expert-review priorities. This makes HantaWatch a practical decision-support layer for decentralized hantavirus surveillance—augmenting human expertise rather than replacing laboratory or public-health interpretation.