Prokaryote phylogeny meets taxonomy: An exhaustive comparison of composition vector trees with systematic bacteriology

Abstract

We perform an exhaustive, taxon by taxon, comparison of the branchings in the composition vector trees (CVTrees) inferred from 432 prokaryotic genomes available on 31 December 2006, with the bacteriologists' taxonomy--primarily the latest online Outline of the Bergey's Manual of Systematic Bacteriology. The CVTree phylogeny agrees very well with the Bergey's taxonomy in majority of fine branchings and overall structures. At the same time most of the differences between the trees and the Manual have been known to biologists to some extent and may hint at taxonomic revisions. Instead of demonstrating the overwhelming agreement this paper puts emphasis on the biological implications of the differences.

Efficient tools for comparative substring analysis

“…The networks calculated using KL(P obs S,t , P exp S,t ) do not show any improvement, as the organisms from the T. phyla are collected in essentially 2 clusters. However, the networks calculated using surp(P obs S,t , P exp S,t ), though less resolved, are much more stable and seem to essentially agree with trees or networks generated from existing tools Gao et al, 2004). They correctly place organisms according to their phyla while being consistent with the previous trees on organisms present to the previous dataset.…”

“…In view of the results obtained in QI et al (2004) and Gao et al (2004), there is a good reason to expect that, looking for triples a, t, b for which P obs S,t (b, a) differs substantially from P exp S,t (b, a) -or just for test strings t for which the probability distribution P obs S,t differs substantially from the probability distribution P exp S,t (as measured, e.g., by the Kullback-Leibler distance; Cover and Thomas, 1991)-might help in many ways in comparative genome (and even meta-genome) analysis including the reconstruction of phylogenetic trees relying on composition-based distances. And, as we will see in the next section, insights obtained and ideas introduced and well-worked out already by the suffix-tree community, enable us to compute all of these probability distribution simultaneously for all test-strings t in linear time (relative to total input length).…”

“…Further, to reduce the analysis of the family S of N sequences to one-sequence analysis, we follow standard procedures by extending the alphabet A by a p unctuation character "$", that is assumed not to be contained in A, thus forming the new alphabetÂ:=A ∪ {$} which allows us to define the "long" sequence Next, we follow QI et al (2004) and Gao et al (2004) by focusing (rather than on just the bare frequencies of the test strings t in S) on the relative frequencies of one-letter extensions of the form t • a, b • t and b • t • a of our test string t(a, b ∈Â). To this end, we define, for any family S of words as above and any substring t = t 1 t 2 .…”

“…In this paper, we continue investigating Composition Vectors, a conceptual tool for alignment-free comparison of sequence families that has been crucial for the work presented in QI et al (2004) and Gao et al (2004).…”

Prediction of success for polymerase chain reactions using the Markov maximal order model and support vector machine

“…The first usage of k-word counts for biological sequence comparison was implemented by Blaisdell (1986), and then it was soon followed by a number of attempts for developing k-word models (Burge et al, 1992;Karlin and Ladunga, 1994;Karlin and Burge, 1995;Rocha et al, 1998;Chen et al, 2013, Xu et al, 2013Qiu et al, 2014;Hao et al, 2003;Qi et al, 2004;Gao et al, 2007;Wang et al, 2009;Pride et al, 2003;Wu et al, 2007;He and Xia, 2007;Dai and Wang, 2008;Zhang et al, 2011;Yu et al, 2012;Ding et al, 2013;Yang and Wang, 2013;Li et al, 2014). However, most of these methods neglected the relationships among word counts in a sequence.…”