|PELZEL-MCCLUSKEY, ANGELA - Animal And Plant Health Inspection Service (APHIS)|
|BURRUSS, N. DYLAN - New Mexico State University|
|Schrader, Theodore - Scott|
|LOMBARD, JASON - Animal And Plant Health Inspection Service (APHIS)|
Submitted to: Ecosphere
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/6/2020
Publication Date: 6/21/2020
Citation: Peters, D.C., McVey, D.S., Elias, E.H., Pelzel-McCluskey, A.M., Derner, J.D., Burruss, N., Schrader, T.S., Yao, J., Pauszek, S.J., Lombard, J., Rodriguez, L.L. 2020. Big data-model integration and AI for vector-borne disease prediction. Ecosphere. 11:1-20. https://doi.org/10.1002/ecs2.3157.
Interpretive Summary: We applied a framework that coupled human and machine learning with big data-model integration to inform the development of early warning strategies (EWS) for vector-borne diseases. Our trans-disciplinary expertise was used to identify, geo-reference, and integrate environmental variables at multiple scales across the geographical extent of a disease. Statistical analyses determined important variables explaining patterns in disease occurrence. We used this framework to explain patterns in occurrence of vesicular stomatitus (VS), the most common vesicular disease affecting livestock in the Americas. Our results show that viral phylogeny was less important than environmental variables that favored different vectors at landscape to regional scales. At local scales, VS occurrence was related to conditions that can be monitored (rainfall, temperature) or managed (vegetation) as part of EWS. This framework can be applied to other vector-borne diseases where big data-model integration is needed for more complete understanding and management.
Technical Abstract: Predicting the drivers of incursion and expansion of vector-borne diseases as part of early-warning strategies (EWS) is a major challenge for geographically extensive diseases where spread is mediated by spatial heterogeneity in climate and other environmental drivers. Geospatial data on these environmental drivers are increasingly available affording opportunities for application to a predictive disease ecology paradigm provided the data can be synthesized and harmonized with fine-scale, highly resolved data on vector and host responses to their environment. Here, we apply a multi-scale big data–model integration approach using human-guided machine learning to objectively evaluate the importance of a large suite of spatially distributed environmental variables (>400) to develop EWS for vesicular stomatitis (VS), a common viral vectorborne vesicular disease affecting livestock throughout the Americas. Two temporally and phylogenetically distinct events were used to develop disease occurrence–environment relationships in incursion (2004) and expansion years (2005), and then to test those relationships (2014, 2015) at two scales: (1) local and (2) landscape to regional. Our results show that VS occurrence at a local scale of individual landowners was related to conditions that can be monitored (rainfall, temperatures, streamflow) or modified (vegetation). On-site green vegetation during the month of occurrence and higher rainfall four months prior combined with either cool daytime (expansion) or nighttime (incursion) temperatures one month prior were indicators of VS occurrence. Distance to running water (incursion) and host density based on neighboring ranches (expansion) with infected animals were also important in individual years. At landscape-to-regional scales, conditions that favor specific VSV biological vectors were indicated, either black flies in incursion years or biting midges in expansion years. Changes in viral genetic lineage were less important to patterns in VS occurrence than factors affecting the host–vector–environment interactions. In combination with our onset map based on latitude, elevation, and long-term annual precipitation, this year- and scale-specific information can be used to develop strategies to minimize effects of future VS events. This big data approach coupled with expert knowledge and machine learning can be applied to other emerging diseases for improvement in understanding, prediction, and management of vector-borne diseases.