Leveraging recurrent graph neural networks to improve geospatial estimation of equine West Nile Virus outbreaks

West Nile Virus (WNV) poses a significant public health threat to humans and livestock. Its transmission dynamics are influenced by multiple environmental factors that are related to disease vector dynamics, making it difficult to predict cases in both space and time.

We applied a geospatial recurrent graph neural network artificial intelligence model to identify key factors linked to WNV occurrence in horses and evaluated the ability of the model to estimate county‐level WNV occurrence within the southern climate region of the United States, with the aim of enhancing understanding of high‐risk areas for targeted intervention strategies.

The model achieved an area under the receiver operating characteristic curve of 0.990 for the held‐out 2012 outbreak year and 0.995 for the pooled 2018–2019 evaluation period. Influential predictors were linked to mosquito vector ecology and included climatic, vegetation and land cover variables. Among county‐weeks classified as very high risk, mean monthly precipitation was mm, mean monthly temperature was , and drought was generally absent.

Synthesis and applications . These findings contribute to ongoing efforts to mitigate the impact of WNV by refining our understanding of the spatial determinants of virus transmission, informing evidence‐based interventions and enhancing preparedness and response strategies. This framework supports targeted intervention strategies, such as conducting vaccine campaigns prior to optimal WNV transmission conditions or prioritizing locations for mosquito control.