Abstract
Ebola and other filoviruses pose significant public health and conservation threats by causing high mortality in primates, including humans. Preventing future outbreaks of ebolavirus depends on identifying wildlife reservoirs, but extraordinarily high biodiversity of potential hosts in temporally dynamic environments of equatorial Africa contributes to sporadic, unpredictable outbreaks that have hampered efforts to identify wild reservoirs for nearly 40 years. Using a machine learning algorithm, generalized boosted regression, we characterize potential filovirus-positive bat species with estimated 87% accuracy. Our model produces two specific outputs with immediate utility for guiding filovirus surveillance in the wild. First, we report a profile of intrinsic traits that discriminates hosts from non-hosts, providing a biological caricature of a filovirus-positive bat species. This profile emphasizes traits describing adult and neonate body sizes and rates of reproductive fitness, as well as species' geographic range overlap with regions of high mammalian diversity. Second, we identify several bat species ranked most likely to be filovirus-positive on the basis of intrinsic trait similarity with known filovirus-positive bats. New bat species predicted to be positive for filoviruses are widely distributed outside of equatorial Africa, with a majority of species overlapping in Southeast Asia. Taken together, these results spotlight several potential host species and geographical regions as high-probability targets for future filovirus surveillance.