Predicting the global mammalian viral sharing network using phylogeography

Gregory F Albery; Evan A Eskew; Noam Ross; Kevin J Olival

lumn (Species 2) of the 569 sharing matrix. Using the paraPen specification in mgcv, these random effects were We validated the predicted network by comparing it to sharing patterns in the Enhanced 633 Infectious Diseases Database (EID2) 29 . We eliminated species pairs that were in our training 634 data and identified whether species pairs that shared viruses in EID2 were more likely to 635 share viruses in our predicted network than species pairs that did not. In addition, we 636 investigated whether species that were shown to host zoonoses in our training dataset were 637 more highly-connected in the predicted network. Finally, we investigated whether species 638 that were present in only EID2, in only our training data, or in both were more highly-639 connected in our predicted network than species that did not appear in either dataset and were 640 therefore taken to have not been observed hosting a virus. The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/732255 doi: bioRxiv preprint viruses with the remaining species, our model was taken to be useful for predicting patterns 648 of mammal sharing based on known host distributions. The mean ranking of the focal hosts 649 across each prediction iteration was used as a measure of "predictability" for each virus. We 650 carried out this process for the 250 viruses with more than one known host with associated 651 geographic and phylogenetic data and then on the 109 such viruses in the EID2 data. 652 Once the predictability of each virus was calculated, we fitted a linear mixed model 653 examining log10(mean focal host rank) as an inverse measure of predictability (higher rank 654 corresponds to decreased predictability) for each virus. We added mean phylogenetic host 655 similarity as a fixed effect and viral family as a random effect to quantify how viral 656 phylogeny affected predictability. We included additional viral traits in the model, including: 657 cytoplasmic replication (0/1); segmentation (0/1); vector-borne transmission (0/1); double-or 658 single-strandedness; DNA or RNA; enveloped or non-enveloped; or zoonotic ability (0/1 for 659 whether the virus was associated with humans in our dataset). Figure SI4 : Mammal species that were observed with at least one virus in the training dataset or the 690 EID2 dataset had higher degree centrality (link number) in our predicted network. This figure displays 691 the raw data that are displayed in Figure 2C in the main text, but without being scaled within orders. 692 693 . CC-BY-NC-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/732255 doi: bioRxiv preprint The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/732255 doi: bioRxiv preprint Figure SI6 : Scaling of degree centrality (link numbers) followed a power law when looking within-701