Fig 1.
Traits predictive of carnivore RABV reservoir status based on the conservative criteria.
(A) Trait importance (measured based on model error change after permutation) and (B)-(E) partial dependence plots relating RABV reservoir status (the log-odds scale of being a reservoir for RABV) with the four most predictive carnivore traits based on the conservative criteria. In panel (A), PCoA1-3 refers to the first to third principal coordinates of species phylogenetic relatedness. In panels (B)-(D), the red line represents the mean prediction across all species. The grey lines are the Individual Conditional Expectation (ICE) curves, which illustrate the predictive change in each species being a reservoir for RABV as each feature changes. The tick marks along the x-axis represent the deciles of each trait value included in model training. The median age at sexual maturity was ~2 years and the median litter size was 2.35.
Fig 2.
Illustrative examples of (A) a predicted reservoir: The least weasel (Mustela nivalis); (B) a non-reservoir: The lion (Panthera leo); and (C) a known reservoir: The red fox (Vulpes vulpes).
Bars denote trait importance based on the Shapley value (phi). Positive Shapley values indicate that predictors are increasing the likelihood that the outcome is positive (i.e., the likelihood a species is a reservoir for RABV), and negative Shapley values indicate that predictors are increasing the likelihood that the outcome is negative (i.e., the likelihood a species is not a reservoir for RABV). Values next to each trait represent the trait measure for each one of the three species (e.g., obtained from the PanTHERIA database). Least weasel and red fox photos were obtained from Wikimedia Commons (https://commons.wikimedia.org/) and the lion photo was the authors’ contribution. The text at the top of each image represents the reservoir status of each species based on the literature (black) and GBM models (blue).
Table 1.
Carnivore species predicted to be reservoirs for RABV based on the conservative criteria.
Since there is inherent variation when performing permutations, species with Shapley values close to zero (especially those < 0.1) should be considered with caution.
Fig 3.
Distribution of carnivore RABV reservoirs identified by the conservative model ((A) and (C)) and the liberal model ((B) and (D)).
Panel (A) and (B) represent known reservoirs, and panel (C) and (D) predicted reservoirs. The maps show areas with high (red), moderate (orange), and low (yellow) number of carnivore reservoirs for RABV. Grey histograms represent the richness level (i.e., the number of reservoir species in each pixel).
Fig 4.
Traits predictive of bat RABV reservoir status based on the conservative criteria.
(A) Trait importance (measured based on model error change after permutation) and (B)-(D) partial dependence plots relating reservoir status (the log-odds scale of being a reservoir for RABV) with the three most predictive bat traits based on the conservative criteria. In panel (A), PCoA1-4 refers to the first to fourth principal coordinates of species phylogenetic relatedness. In panels (B)-(C), the red line represents the mean prediction across all species. The grey lines are the Individual Conditional Expectation (ICE) curves, which illustrate the predictive change in each species being a reservoir for RABV as each feature changes. The tick marks along the x-axis represent the deciles of each trait values included in model training.
Fig 5.
Illustrative examples of (A) a predictive reservoir: The long-legged myotis (Myotis volans); (B) a non-reservoir: The hairy fruit-eating bat (Artibeus hirsutus); and (C) a known reservoir: The vampire bat (Desmodus rotundus).
Bars denote trait importance based on the Shapley value (phi). Positive Shapley values indicate that predictors are increasing the likelihood that the outcome is positive (i.e., a species is a reservoir for RABV), and negative Shapley values indicate that predictors are increasing the likelihood that the outcome is negative (i.e., a species is not a reservoir for RABV). Values next to each trait represent the trait measure for each one of the three species (obtained from the PanTHERIA database). All photos were obtained from Wikimedia Commons (https://commons.wikimedia.org/). The text at the top of each image represents the reservoir status of each species based on the literature (black) and GBM models (blue).
Table 2.
Bat species predicted to be RABV reservoirs based on the conservative criteria.
Since there is inherent variation when performing permutations, species with Shapley values close to zero (especially those < 0.1) should be considered with caution.
Fig 6.
Geographic distribution of bat RABV reservoirs identified by the conservative model ((A) and (C)) and the liberal model ((B) and (D)).
Panel (A) and (B) represent known reservoirs, and panel (C) and (D) predicted reservoirs. The maps show areas with high (red), moderate (orange), and low (yellow) number of bat RABV reservoirs. Grey histograms represent the richness level (i.e., the number of reservoir species in each pixel).