Early detection of animal species capable of hosting novel viruses is critical for avoiding future health catastrophes. This is the setting for a study undertaken by Sapienza University of Rome's Departments of Physics, Biology, and Biotechnology "Charles Darwin" and published in Nature Communications. The study used an advanced artificial intelligence algorithm to detect relationships between virus families and mammals. Currently, less than 1% of animal viruses capable of infecting humans are identified; pathogens such as SARS, Ebola, and SARS-CoV-2 indicate how the potential of future outbreaks remains high due to huge gaps in worldwide scientific mapping. As the researchers noted, the data now available is frequently incomplete: the absence of reports in a species does not imply immunity, but rather represents a lack of studies in certain geographic locations. To address this bias, the researchers created Dynamic Positive-Unlabeled (DPU) learning, a machine learning technique that can distinguish between the genuine absence of a virus and a lack of data. Field experiments have already confirmed the model's correctness, indicating that the number of interactions between mammals and viruses is far higher than previously thought.