Predicting Mammal Species Suitable for Chagas Disease Spread Using a Machine-Learning Framework.

Abstract

Understanding which mammalian species participate in the sylvatic cycle of Trypanosoma cruzi is vital for clarifying the ecological processes that maintain Chagas disease transmission. Although more than 100 mammal species have been reported as naturally infected, the actual diversity of hosts is much broader. In this study, we employed a trait-based, machine-learning framework to evaluate mammals associated with natural infection. Data concerning life-history traits, morphological characteristics, environmental factors, trophic ecology, habitat use, biogeographic distributions, and spatial overlap with triatomine vectors were compiled to model compatibility with T. cruzi infection using extreme gradient-boosted regression. Our analysis indicates that triatomines overlap with 573 mammal species. While triatomines are considered generalists, they also utilize mammal species with specific traits, such as den-dwelling behavior, medium to large body size, and tolerance to habitat disturbance. This approach offers a structured foundation for identifying potential host species and prioritizing field research in regions that are insufficiently studied. By integrating ecological traits with geographic and behavioral data, the model enhances the understanding of host-parasite interactions in sylvatic Chagas disease and supports targeted surveillance of predicted mammalian hosts.