Fig 1.
Graphical representation of workflow.
(a) Each virus is made up of genes, some of which may match the genes on another virus (Viruses A and B, respectively). We imagine the variation of information between two genes Di as the resistance of a resistor connecting them. Then the total “distance” between the collections of genes in Viruses A and B is the equivalent resistance between the two sides. Only the genes Ai and Bj which match are shown and indexed; the total number of options for choosing one gene from each virus is quite large, and most such choices do not yield a match. These pairs are analogous to open circuits, with infinite resistance, which do not affect the equivalent resistance. (b) K-mer distance is calculated by assessing whether knowing that a k-mer is over-represented in one genome indicates anything about whether it is over-represented in another genome. (c) The k-mer distance is scaled and then combined with the BLAST distance. (d) Classical multidimensional scaling transforms the distance matrix into a Euclidean position matrix. (e) Three-dimensional embedding with t-SNE allows for clustering and interpretation. (f) Density-based clustering with OPTICS yields 78 clusters, and is the basis for further analysis. Although transformations in (c) and (e) rescale distances, the axes of the histograms are all scaled to correspond with one another.
Fig 2.
t-SNE plot of 5,817 viruses from RefSeq, grouped by mutual information of genes in common and 4-mer frequency.
Points are colored according to 78 clusters assigned by density-based clustering in the 3-dimensional t-SNE space. Each cluster is numbered, and cluster numbers correspond to those in Fig 5. Most clusters correspond to one order or family in the ICTV classification.
Fig 3.
t-SNE plot of 5,817 viruses from RefSeq, grouped by mutual information of genes in common and 4-mer frequency.
Points are colored by Baltimore classification: red = dsDNA viruses (Baltimore class I & VII), green = ssDNA viruses (Baltimore class II), blue = dsRNA viruses (Baltimore class III), yellow = ssRNA viruses, positive sense (Baltimore class IV & VI), brown = ssRNA viruses, negative sense (Baltimore class V), and gray = unclassified.
Fig 4.
t-SNE plot of 5,817 viruses from RefSeq, grouped by mutual information of genes in common and 4-mer frequency.
Points are colored by host kingdom: magenta = Archaea, maroon = Eubacteria, sky blue = Fungi, lime green = Plantae, orange = Animalia, and gray = unclassified.
Fig 5.
Dendrogram of 78 clusters of viruses, determined by density-based clustering on the 3-dimensional t-SNE space of viral genomes.
Cluster numbers correspond to numbers in Fig 4. Three numbers are given in parentheses: the number of viruses in each cluster, the number of viruses which are not part of the family but nevertheless are clustered together with it, and the number of viruses in a family which were not included in its cluster. Asterisks after the second number indicate that most of the incorrect viruses are unclassified by ICTV or otherwise correspond weakly with it. In a few cases, one family was split into multiple clusters; in these cases, the clusters were designated “A,” “B,” etc. If any viruses in the family were missing from all such clusters, their count was divided equally between them, leading to the occasional fractional virus count in the third coordinate. Because this dendrogram was developed from hierarchical clustering, branch lengths correspond to how much more similar a given cluster is to itself than to the other viruses.
Fig 6.
Dendrogram of cluster 32, which corresponds to the Flaviviruses.
Of the 44 viruses in this cluster, 19 which are clinically significant in humans are labeled; taxa names for the other 25 are omitted for clarity. This constructed phylogeny suggests that Zika is more genetically similar to the West Nile virus and yellow fever than to the Dengue viruses; the closest Dengue virus is Dengue 4. The two species marked with asterisks are not Flaviviruses; although they were clustered together with them, they are separated from the true Flaviviruses. Eight Flaviviruses make up a single sub-cluster of cluster 16 instead.