Quartet-Based Computations of Internode Certainty Provide Robust Measures of Phylogenetic Incongruence

Abstract

Incongruence, or topological conflict, is prevalent in genome-scale data sets. Internode certainty (IC) and related measures were recently introduced to explicitly quantify the level of incongruence of a given internal branch among a set of phylogenetic trees and complement regular branch support measures (e.g., bootstrap, posterior probability) that instead assess the statistical confidence of inference. Since most phylogenomic studies contain data partitions (e.g., genes) with missing taxa and IC scores stem from the frequencies of bipartitions (or splits) on a set of trees, IC score calculation typically requires adjusting the frequencies of bipartitions from these partial gene trees. However, when the proportion of missing taxa is high, the scores yielded by current approaches that adjust bipartition frequencies in partial gene trees differ substantially from each other and tend to be overestimates. To overcome these issues, we developed three new IC measures based on the frequencies of quartets, which naturally apply to both complete and partial trees. Comparison of our new quartet-based measures to previous bipartition-based measures on simulated data shows that: (1) on complete data sets, both quartet-based and bipartition-based measures yield very similar IC scores; (2) IC scores of quartet-based measures on a given data set with and without missing taxa are more similar than the scores of bipartition-based measures; and (3) quartet-based measures are more robust to the absence of phylogenetic signal and errors in phylogenetic inference than bipartition-based measures. Additionally, the analysis of an empirical mammalian phylogenomic data set using our quartet-based measures reveals the presence of substantial levels of incongruence for numerous internal branches. An efficient open-source implementation of these quartet-based measures is freely available in the program QuartetScores (https://github.com/lutteropp/QuartetScores).

The phylogeny of insects in the data‐driven era

“…Finally, the research field currently faces a challenge in that bootstrap support appears unsuited as a measure of node support for genome‐scale data (Townsend & Naylor, ; Kumar et al ., ; Salichos & Rokas, ; Freitas et al ., ). Here I opt for a conservative measure of node support in the form of IC calculated from gene tree variation, although these measures are under active development and can vary considerably based on approach used to adjust for taxon completeness (Zhou et al ., ). Other options that might allow us to move forward in node assessment could be related measures of gene tree conflict (Salichos et al ., ; Zhou et al ., ), concatenation analyses (Narechania et al ., ) and four‐cluster likelihood mapping (Bank et al ., ; Johnson et al ., ).…”

“…Here I opt for a conservative measure of node support in the form of IC calculated from gene tree variation, although these measures are under active development and can vary considerably based on approach used to adjust for taxon completeness (Zhou et al ., ). Other options that might allow us to move forward in node assessment could be related measures of gene tree conflict (Salichos et al ., ; Zhou et al ., ), concatenation analyses (Narechania et al ., ) and four‐cluster likelihood mapping (Bank et al ., ; Johnson et al ., ). At the very least, it is necessary to assess sensitivity of the phylogenetic results to the way in which the matrix is constructed (Whelan et al ., ; Garrison et al ., ; Cunha & Giribet, ), including the use of matrices containing orthologues that are more complete for taxa relevant to the study question (decisive datasets; Dell'Ampio et al ., ; Parks et al ., ).…”

Disentangling Sources of Gene Tree Discordance in Phylogenomic Datasets: Testing Ancient Hybridizations in Amaranthaceae s.l

“…2) Data processing: care should be taken in data cleaning, partitioning (e.g., nuclear vs. plastid), and using orthology inference methods that explicitly address paralogy issues (e.g., treebased orthology inference and synteny information). 3) Species tree inference: select species tree methods that accommodate the dataset size and data type (e.g., ASTRAL for gene tree-based inferences or SVDquartet [Chifman and Kubatko 2014] for SNP-based inferences), followed by visualization of phylogenetic conflict using tools such as the pie charts (e.g., PhyParts) and quartet-based tools (e.g., Quartet Sampling; Quadripartition Internode Certainty [Zhou et al 2020]; Concordance Factors [Minh et al 2020]). 4) Assessing hybridization: if phylogenetic conflict cannot be explained by processes like ILS, phylogenetic species network analyses (e.g., PhyloNet) reduced taxon sampling can be applied to test hybridization hypotheses given results in step 3; 5) Hypothesis testing: additional tests can be performed given the results of recommendation 3 and 4 and depending on the scenario.…”

A deeper meaning for shallow‐level phylogenomic studies: nested anchored hybrid enrichment offers great promise for resolving the tiger moth tree of life (Lepidoptera: Erebidae: Arctiinae)

“…Low phylogenetic support can be caused by poor phylogenetic signal and/or conflicting signals among the loci used for species tree inference. To test the effects of conflicting phylogenetic signal within the data set, we calculated the quadripartition internode certainty score (QP-IC; Zhou et al, 2019) of internodes within the species trees. We calculated the QP-IC scores using the program QuartetScores, a quartet-based measure for examining incongruence within a set of phylogenetic trees (available at https://github.com/lutteropp/QuartetScores).…”

“…The average QP-IC score for the concatenated maximum likelihood tree of all taxa was 0.329 (File S10). Quartet-based IC scores can be open to interpretation (Zhou et al, 2019), making it difficult to objectively evaluate these scores, however, this result may be indicative of moderate levels of phylogenetic incongruence of the gene trees relative to the species tree. The only negative scores, indicating an alternative topology was more prevalent among the gene trees, corresponded to branches descended from the root and may be a technical artifact.…”