Viral taxonomy derived from evolutionary genome relationships.

We describe a new genome alignment-based model for understanding the diversity of viruses based on evolutionary genetic relationships. This approach uses information theory and a physical model to determine the information shared by the genes in two genomes. Pairwise comparisons of genes from the viruses are created from alignments using NCBI BLAST, and their match scores are combined to produce a metric between genomes, which is in turn used to determine a global classification using the 5,817 viruses on RefSeq. In cases where there is no measurable alignment between any genes, the method falls back to a coarser measure of genome relationship: the mutual information of 4-mer frequency. This results in a principled model which depends only on the genome sequence, which captures many interesting relationships between viral families, and which creates clusters which correlate well with both the Baltimore and ICTV classifications. The incremental computational cost of classifying a novel virus is low and therefore newly discovered viruses can be quickly identified and classified. The model goes beyond alignment-free classifications by producing a full phylogeny similar to those constructed by virologists using qualitative features, while relying only on objective genes. These results bolster the case for mathematical models in microbiology which can characterize organisms using only their genetic material and provide an independent check for phylogenies constructed by humans, considerably faster and more cheaply than less modern approaches.