Network embedding unveils the hidden interactions in the mammalian virome

• DOI: 10.1016/j.patter.2023.100738
• Publication Date: 2023-04-24T16:23:15Z
• Authors: Timothée Poisot; Marie-Andrée Ouellet; Nardus Mollentze; Maxwell J. Farrell; Daniel J. Becker; Liam Brierley; Gregory F. Albery; Rory J. Gibb; Stephanie N. Seifert; Colin J. Carlson
• Keywords: Human virome

Predicting host-virus interactions is fundamentally a network science problem. We develop method for bipartite prediction that combines recommender system (linear filtering) with an imputation algorithm based on low-rank graph embedding. test this by applying it to global database of mammal-virus and thus show makes biologically plausible predictions are robust data biases. find the mammalian virome under-characterized anywhere in world. suggest future virus discovery efforts could prioritize Amazon Basin (for its unique coevolutionary assemblages) sub-Saharan Africa poorly characterized zoonotic reservoirs). Graph embedding imputed improves human infection from viral genome features, providing shortlist priorities laboratory studies surveillance. Overall, our study indicates structure contains large amount information recoverable, provides new insights into fundamental biology disease emergence.