Predicting larval tick burden on white-footed mice with an artificial neural network

Abstract

White-footed mice are important hosts for immature blacklegged ticks (Ixodes scapularis) and the most competent reservoir hosts for several tick-borne pathogens, including the agent of Lyme disease, in eastern North America. The distribution of larval ticks on individual mice tends to be highly heterogeneous, potentially resulting in few individual hosts causing the majority of host-to-tick transmission events. In this study, we created an artificial neural network (ANN) model using a 20 year data set from Millbrook, NY, to understand which attributes of mice or the environment predict high larval burden. Furthermore, we performed a sensitivity analysis to explore the importance of, and interactions between, the most influential attributes. Our analysis indicated that highest larval burden is predicted in warmer and drier than average years when host abundance is low, and that climatic conditions and host density are far more important in predicting larval burden than traits of individual mice, a finding that could have human health implications within the context of a warming climate. Practically, our results suggest that instead of basing tick-control treatments on particular attributes of hosts, treatments should be targeted based on climate factors. Additionally, our results highlight the importance of including variable interactions in models aiming to predict vector (tick) aggregation, and, most broadly, demonstrate the utility of ANNs in understanding aggregation of ticks and other vectors.