Machine learning algorithms for the evaluation of risk by tick-borne pathogens in Europe

Abstract

Background Tick-borne pathogens pose a major threat to human health worldwide. Understanding the epidemiology of tick-borne diseases to reduce their impact on human health requires models covering large geographic areas and considering both the abiotic traits that affect tick presence, as well as the vertebrates used as hosts, vegetation, and land use. Herein, we integrated the public information available for Europe regarding the variables that may affect habitat suitability for ticks and hosts and tested five machine learning algorithms (MLA) for predicting the distribution of four prominent tick species across Europe. Materials and methods A grid of cells 20 km in diameter was prepared to cover the entire territory, containing data on vegetation, points of water, habitat fragmentation, forest density, grass extension, or imperviousness, with information on temperature and water deficit. The distribution of the hosts (162 species) was modelled and included in the dataset. We used five MLA, namely, Random Forest, Neural Networks, Naive Bayes, Gradient Boosting, and AdaBoost, trained with reliable coordinates for Ixodes ricinus , Dermacentor reticulatus , Dermacentor marginatus , and Hyalomma marginatum in Europe. Results Both Random Forest and Gradient Boosting best predicted ticks and host environmental niches. Our results demonstrate that MLA can identify trait-matching combinations of environmental niches. The inclusion of land cover and land use variables has a superior capacity for predicting areas suitable for ticks, compared to classic methods based on the use of climate data alone. Conclusions Flexible MLA-driven models may offer several advantages over traditional models. We anticipate that these results may be extrapolated to other regions and combinations of tick-vertebrates. These results highlight the potential of MLA for inference in ecology and provide a background for the evolution of a completely automatized tool to calculate the seasonality of ticks for early warning systems aimed at preventing tick-borne diseases.