Abstract
Bats and the viruses they host are central to understanding zoonotic disease emergence. Eurasia forms a vast, interconnected habitat for many bat species, and extensive migrations suggest that these populations function as linked transmission networks rather than isolated groups. Public databases such as DBatVir/ZOVER2.0 enable large-scale comparative analyses across bat hosts and viruses. In this study, we examined the Eurasian subset to describe geographic and taxonomic patterns in bat-associated viruses and compare machine-learning approaches with zero-inflated count models. The dataset included 9,442 cleaned records, with 7,839 from Asia and 1,603 from Europe. These records represented 28 viral families and 153 distinct bat species after taxonomic harmonization. Data distribution was highly uneven, with the ten most frequently reported species accounting for 47.8% of all records. Frequently recorded hosts included Scotophilus kuhlii, Rhinolophus sinicus, and Miniopterus schreibersii. Coronaviridae was the most commonly reported viral family (5,526 records), followed by Astroviridae and Rhabdoviridae. Asia showed a strong dominance of coronavirus records, whereas Europe exhibited a more diverse viral profile with relatively more rhabdoviruses. Machine-learning models achieved moderate accuracy (0.73-0.76), but macro-F1 scores remained low (0.37-0.42), indicating weaker performance for rare viral families. A zero-inflated Poisson model estimated significantly lower expected counts in Europe compared to Asia, yielding an incidence rate ratio of 0.245 (95% CI 0.232-0.259). Overall, the dataset is suitable for identifying broad geographic trends but less reliable for detailed analysis of rare viruses. These differences likely reflect ecological variation, host susceptibility, exposure risk, and uneven surveillance effort across regions and time.