New Methods to Calculate Concordance Factors for Phylogenomic Datasets.
Minh BQ, Hahn MW, Lanfear R. · 2020
Read New Methods to Calculate Concordance Factors for Phylogenomic Datasets. by Minh BQ, Hahn MW, Lanfear R. (2020) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc7475031.txt.
Introduction
Minh, Hahn, and Lanfear give new methods to measure agreement among gene trees in phylogenomic data. Two scores do the work. The gene concordance factor shows how many decisive gene trees hold a given branch.
The site concordance factor is new here. It shows how many decisive DNA sites back a branch. Both scores give a full picture of disagreement that classic support values miss. The package handles uneven species coverage across gene trees, a common flaw in big data sets. It runs inside the free IQ-TREE software, with a tutorial for users. Rooted tree cases get their own treatment. The authors show how to read the scores on real data. The tools help teams judge which branches of an evolution tree are solid.
Catalog record
| CREATOR | Minh BQ, Hahn MW, Lanfear R. |
|---|---|
| YEAR | 2020 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
New Methods to Calculate Concordance Factors for Phylogenomic Datasets By Bui Quang Minh, Matthew W Hahn, Robert Lanfear, Michael Rosenberg Published in: Molecular Biology and Evolution DOI: 10.1093/molbev/msaa106 ABSTRACT Abstract We implement two measures for quantifying genealogical concordance in phylogenomic data sets: the gene concordance factor (gCF) and the novel site concordance factor (sCF). For every branch of a reference tree, gCF is defined as the percentage of “decisive” gene trees containing that branch. This measure is already in wide usage, but here we introduce a package that calculates it while accounting for variable taxon coverage among gene trees. sCF is a new measure defined as the percentage of decisive sites supporting a branch in the reference tree. gCF and sCF complement classical measures of branch support in phylogenetics by providing a full description of underlying disagreement among loci and sites. An easy to use implementation and tutorial is freely available in the IQ-TREE software package (http://www.iqtree.org/doc/Concordance-Factor, last accessed May 13, 2020). Keywords: phylogenetic inference, concordance factor, phylogenomics Measures of branch support such as the bootstrap (Felsenstein 1985; Minh et al. 2013) and Bayesian posterior probability (Ronquist and Huelsenbeck 2003) are important for making robust inferences from phylogenetic trees. However, although these measures provide useful information about the statistical support for a given branch, neither captures the topological variation present in the underlying data (Kumar et al. 2012). A complementary approach involves calculating the fraction of loci consistent with a particular branch, thus capturing underlying agreement and disagreement in the data. Various approaches have been suggested, including the gene concordance factor (gCF) (Baum 2007)—sometimes also referred to as the “gene support frequency” (Gadagkar et al. 2005; Salichos and Rokas 2013)—and internode certainty (Salichos and Rokas 2013), with the former being the most widely used. The gCF describes for each branch in a reference tree the proportion of inferred single-locus trees that contain that branch. Although intuitive, the gCF suffers from three limitations: first, there are few software implementations (Ané et al. 2007); second, although issues with incomplete taxon sampling of gene trees have been addressed by some authors (Smith et al. 2015; Kobert et al. 2016), these fixes are not available in implementations that allow for the calculation of gCF values; and third, low gCF values are hard to interpret because they may result from strongly supported discordance among individual gene trees or weak phylogenetic signal in individual loci. Here, we resolve these issues by implementing the calculation of gCF while accounting for unequal taxon sampling in the popular IQ-TREE package (Nguyen et al. 2015; Minh et al. 2020), and by introducing the site concordance factor (sCF), a measure that estimates concordance at the level of individual sites. We argue and show by detailed analyses of ten empirical datasets that concordance factors complement commonly used metrics like the bootstrap by providing additional information and insights about topological variation. GENE CONCORDANCE FACTOR For a given unrooted bifurcating reference tree, T (e.g., an estimate of the species tree), and a set of unrooted bifurcating input trees, S = T1,…,Tn (e.g., individual gene trees), we can calculate the gCF for every internal branch x in T. Each gene tree, Ti, must contain a subset of the taxa in T but need not include all of the same taxa. Each internal branch x of T is associated with four clades of T representing four taxon subsets, A, B, C, and D, such that the bipartition A∪B|C∪D corresponds to the branch x (fig. 1). The reference tree T contains branch x by definition, but branch x may or may not be present in a gene tree of the same taxa (or a subset of them). Let Ai, Bi, Ci, and Di be the set of taxa in A, B, C, and D that are also present in Ti. We say Ti is decisive for x if Ai, Bi, Ci, and Di are nonempty (we use “decisive” in the sense of Sanderson et al. [2011], but note for clarity that Dell’Ampio et al. [2014] use this term in a different way): that is, a gene tree is decisive as long as it contains at least one of the taxa in each of the taxon subsets A, B, C, and D. A decisive gene tree can potentially contain the branch x. We say Ti is concordant with x if Ti is decisive for x and the bipartition Ai∪Bi|Ci∪Di corresponds to a branch in Ti. In other words, a gene tree is concordant with the reference tree if it could have contained branch x (i.e., it is decisive) and it does contain branch x. The gCF for branch x is now defined as [formula: gCFx=|{i:Ti is concordant with x}|{i:Ti is decisive for x}.] [Figure: Schematic view of a bifurcating tree at an internal branch x with four surrounding subtrees containing the taxon sets A, B, C, and D. Branch x corresponds to the bipartition A∪B|C∪D. In our definition, branch x is always present in the reference tree, but it may or may not be present in each of the input trees. By applying the two nearest neighbor interchanges (denoted as arrows), one can produce the two other trees that contain taxon sets A, B, C, and D, but with internal branches y and z.] In other words, the gCF is the proportion of input trees decisive for x that are concordant with x; that is, the proportion of all input trees that could have contained x that do contain x. We note that gCF is also used to calculate internode certainty and related measures (Salichos et al. 2014; Kobert et al. 2016). To further understand concordant and discordant gene trees, we categorize the discordant input trees (i.e., those that do not contain branch x) into three groups and calculate three discordance factors: gDF1, gDF2, and gDFP. The first two groups are obtained by applying a nearest neighbor interchange around branch x to result in two alternative topologies (fig. 1) with two branches not appearing in T: y=A∪C|B∪D and z=A∪D|B∪C. Accordingly, we then calculate discordance factors as gCFy=gDF1x and gCFz=gDF2x. Some input trees may not contain any of the branches x, y, or z. This will occur for input trees in which one or more of clades A, B, C, or D are not monophyletic. We define the proportion of decisive input trees that fall into this category as gDFPx, where the “P” stands for “paraphyly.” The four proportions gCFx, gDF1x, gDF2x, and gDFPx will sum to 1, as every decisive input tree must be included in one of the four categories. SITE CONCORDANCE FACTOR To calculate the sCF for branch x, we randomly sample m quartets of taxa q={a, b, c, d}, where a, b, c, and d are in A, B, C, and D, respectively. For each quartet q, we examine the subalignment of taxa a, b, c, and d. For every site j in this alignment, we call jdecisive for x if the characters aj, bj, cj, and dj are all present and j is parsimony informative when restricted to this quartet of taxa. Decisive sites can be concordant or discordant with x. We say that site j is concordant with x if aj=bj≠cj=dj (i.e., j