Extended Data Fig. 9: Predictions of virus-host associations by various bioinformatics approaches largely identify unique interactions within virus-host interaction space.

a, Bioinformatics approaches identifying viral populations predicted to infected the observed metagenome assembled genomes (MAGs). Three different approaches were applied to infer viral populations that infect MAGs; Markov model-based method (WIsH, blue), CRISPER spacer homology match (CRISPR, red) and tRNA homology matches (tRNA, yellow). Numbers within each non-overlapping shaded region show how many MAGs were uniquely predicted as hosts with each method. MAGs predicted as hosts from multiple different methods are found within the overlapping shaded region (for example 2 of the same MAGs were predicted as hosts by WIsH and CRISPR in the red and blue overlapping region). Numbers in parentheses indicate how many of the shared predictions match the same viral population. In all cases, none of the viral populations predicted to infect MAGs were identical between methods. b, Bioinformatics approaches identifying host taxonomy for observed viral populations. Three different approaches were applied to infer host taxonomy; Markov model-based method (WIsH, blue), RNR homology match (RNR, red) and tRNA homology matches (tRNA, yellow). Numbers within each non-overlapping shaded region show how many viral population predictions were unique for each method. Viral populations with hosts predicted from multiple different methods are found within the overlapping shaded region. Numbers in parentheses indicate how many overlapping predictions match at the genus (first) and phylum (second) level. For example, 36 of the same viral populations had host taxonomy predicted by WIsH and RNR (red and blue overlapping region). However, while there were 36 shared predictions, only two of these host predictions were concordant at the genus or phylum level (5.6%).