Recapitulating phylogenies using k-mers: from trees to networks

Abstract

Ernst Haeckel based his landmark Tree of Life on the supposed ontogenic recapitulation of phylogeny, i.e. that successive embryonic stages during the development of an organism re-trace the morphological forms of its ancestors over the course of evolution. Much of this idea has since been discredited. Today, phylogenies are often based on families of molecular sequences. The standard approach starts with a multiple sequence alignment, in which the sequences are arranged relative to each other in a way that maximises a measure of similarity position-by-position along their entire length. A tree (or sometimes a network) is then inferred. Rigorous multiple sequence alignment is computationally demanding, and evolutionary processes that shape the genomes of many microbes (bacteria, archaea and some morphologically simple eukaryotes) can add further complications. In particular, recombination, genome rearrangement and lateral genetic transfer undermine the assumptions that underlie multiple sequence alignment, and imply that a tree-like structure may be too simplistic. Here, using genome sequences of 143 bacterial and archaeal genomes, we construct a network of phylogenetic relatedness based on the number of shared k-mers (subsequences at fixed length k). Our findings suggest that the network captures not only key aspects of microbial genome evolution as inferred from a tree, but also features that are not treelike. The method is highly scalable, allowing for investigation of genome evolution across a large number of genomes. Instead of using specific regions or sequences from genome sequences, or indeed Haeckel’s idea of ontogeny, we argue that genome phylogenies can be inferred using k-mers from whole-genome sequences. Representing these networks dynamically allows biological questions of interest to be formulated and addressed quickly and in a visually intuitive manner.

Robust Inference of Genetic Exchange Communities from Microbial Genomes Using TF-IDF

“…Computational simulations and empirical studies demonstrate that approaches based on k -mer count can support the scalable inference of phylogenies (Chan et al, 2014; Bernard et al, 2016a,b) and identify regions of lateral transfer within a dataset (Cong et al, 2016a,b). Parameters including k can be adjusted to minimize the effects of sequence divergence and genome rearrangement.…”

K-mer similarity, networks of microbial genomes and taxonomic rank

“…# distance using 25-mers (see Methods). For each pair of genomes a and b we transformed $ %& into a similarity value ' %& and generated a similarity network, following our earlier approach 18 ; we consider this network to depict phylogenetic relatedness among these genomes, i.e. to be a phylogenomic network.…”

“…cliques, which reflect evolutionary events and dynamics of microbial genomes. For instance, we recovered the progressive separation of the different genomic lineages throughout their evolution 18 and showcased particular relationships between isolates not observed using a classical tree structure 7 .…”

“…As a proof of concept, we previously generated an AF phylogenetic network for 143 bacterial and archaeal genomes 18 , using pairwise distances based on the ! " # statistic.…”

Benchmarking of alignment-free sequence comparison methods

“…In such cases, a non-tree-like network representation of genome evolution is more appropriate. Recent studies [60,61] have demonstrated the scalability and applicability of AF methods to quickly infer networks of relatedness among microbial genomes. Although we did not consider networks in this study, the curated benchmarking data sets can be easily extended to AF phylogenetic analysis beyond a tree-like structure in the future.…”