henhaobio
Sign in

Genome Biology and EvolutionMolecular BiologyPeer-reviewed journal article

The Prevalence and Impact of Model Violations in Phylogenetics

Suha Naser-KhdouriD, Bùi Quang MinhiD, Wenqi ZhangiD, Eric A. StoneiD, Robert LanfeariD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

In phylogenetic inference we commonly use models of substitution which assume that sequence evolution is stationary, reversible and homogeneous (SRH). Although the use of such models is often criticized, the extent of SRH violations and their effects on phylogenetic inference of tree topologies and edge lengths are not well understood. Here, we introduce and apply the maximal matched-pairs tests of homogeneity to assess the scale and impact of SRH model violations on 3,572 partitions from 35 published phylogenetic datasets. We show that many partitions (39.5%) reject the SRH assumptions, and that for most datasets, the topologies of trees inferred from all partitions differ significantly from those inferred using the subset of partitions that do not reject the SRH assumptions. These results suggest that the extent and effects of model violation in phylogenetics may be substantial. They highlight the importance of testing for model violations and possibly excluding partitions that violate models prior to tree reconstruction. They also suggest that further effort in developing models that do not require SRH assumptions could lead to large improvements in the accuracy of phylogenomic inference. The scripts necessary to perform the analysis are available in https://github.com/roblanf/SRHtests, and the new tests we describe are available as a new option in IQ-TREE (http://www.iqtree.org).

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (16 May 2019) · CC BY 4.0

Permutation-based maximal matched-pairs symmetry tests show 40.8% of 3,572 partitions from 35 published phylogenetic datasets reject stationary, reversible, homogeneous evolution, frequently altering inferred tree topologies.

Most empirical phylogenies use stationary, reversible and homogeneous (SRH) substitution models such as the GTR family, but how often real alignments violate these assumptions, and how much that biases topologies, was poorly characterized. The work extends pairwise symmetry tests to whole alignments with a permutation null, surveys violation across diverse empirical data, and adds a partition-filtering option to IQ-TREE.

Key results

  1. ShownSRH violations are common and vary widely across datasets. Overall and marginal symmetry are rejected much more often than internal symmetry, which points to non-stationarity as the more frequent source of bias.

    Of 3,572 partitions, 1,475 (40.8%) failed MaxSymTest, 1,483 (41.5%) MaxSymTestmar and 312 (8.7%) MaxSymTestint; 1,804 (50.5%) failed at least one test. Per-dataset means were 44.9%, 41.8% and 8.2%; failures were most frequent in third codon positions (viral, mitochondrial, nuclear) and plastid intergenic spacers, but not limited to them. · Violation of SRH conditions is common across 35…; Discussion

  2. ShownPartitions that fail the tests carry a distinct topological signal. Trees from the full data look more like trees from failing partitions than from passing ones, and pass versus fail trees are about 71% as dissimilar as random tree pairs.

    For MaxSymTest and MaxSymTestmar, both normalised path-difference (NPD) and quartet distance placed Tall closer to Tfail than to Tpass; mean NPD between Tpass and Tfail under MaxSymTest across 35 datasets was 0.71, where 1 equals randomly drawn tree pairs. · Distance between trees; Model violation has a large influence on tree…; Tables

  3. ShownThe topological differences are often statistically significant when likelihoods are evaluated on the alignment of partitions that pass the tests.

    Weighted Shimodaira-Hasegawa tests on Dpass with 1000 RELL replicates: under MaxSymTest, Tpass rejected Tall in ~48% of datasets and rejected Tfail in ~84% of datasets. · Tree topology tests; Model violation has a large influence on tree…; Tables

  4. ShownThe number of substitutions in a partition significantly predicts test failure but explains only a minority of the variance. This argues that failures reflect more than differences in statistical power.

    Mixed-effects model (lme4 glmer, dataset as random effect): substitution count explained 17% (MaxSymTest), 18% (MaxSymTestmar) and 2% (MaxSymTestint) of pass/fail variation; predictor p<2e-16; ~80% of variation left to other factors. · Correlation between number of substitutions and…; The number of substitutions explains less than…

  5. SuggestedIn a metazoan phylogenomic dataset, partitions that fail MaxSymTest support a different placement of xenacoelomorph lineages and Spiralia relationships than the full or passing data. The full and passing data match the published tree.

    Cannon et al. 2016 data (212 genes, 424 first/second codon partitions; 34% failed MaxSymTest): Tall and Tpass (281 partitions) identical, Xenacoelomorpha sister to Nephrozoa at 100% bootstrap; Tfail (143 partitions) put Xenoturbella sister to Nephrozoa (93%) and Platyzoa with Acoelomorpha (95%). · Model violation affects the internal…; Discussion; Figure legends

Methods, data and reproducibility

Matched-pairs chi-square symmetry tests (overall, marginal, internal) are extended to alignments by taking the maximum pairwise statistic and scoring it against 999 site-permuted alignments (MaxSymTest); IQ-TREE trees from all, passing and failing partitions are compared by path-difference, quartet distance and weighted Shimodaira-Hasegawa tests.

  • 35 published partitioned phylogenetic datasets (nuclear, mitochondrial, plastid and viral genomes; coding, intron, spacer, tRNA, rRNA, UCE loci), 27 to 355 taxa, 699 to 1,079,052 sites · multiple (animals, plants, viruses) · n = 35 datasets; 3,572 partitions
  • Cannon et al. 2016 metazoan dataset of first and second codon positions from 212 genes, used as the Xenacoelomorpha case study · Metazoa plus choanoflagellate outgroups · n = 76 metazoan taxa + 2 outgroups; 424 partitions
Methods & tools
IQ-TREE (new --maxsymtest option), ModelFinder, Ultrafast bootstrap (1000 replicates), MaxSymTest / MaxSymTestmar / MaxSymTestint (permutation-based), BiSymTest (binomial extension, supplementary), Bowker's and Stuart's matched-pairs symmetry tests (MPTS, MPTMS, MPTIS), Weighted Shimodaira-Hasegawa test with RELL resampling, Normalised Path-Difference (NPD), Quartet distance (QD), lme4 glmer (R), MuMIn r.squaredGLMM (R)
Software & environment
Tests exposed in IQ-TREE via --maxsymtest NUM, --symtest-remove-bad, --symtest-type MAR|INT and --symtest-pval NUM (default 0.05); analysis scripts in Python and R (lme4, MuMIn); no software versions stated.
Reproducibility gaps
No version numbers given for IQ-TREE, R or Python packages; dataset-level details (Table 1, extended tables) and BiSymTest results sit in supplementary material not included in this text.
Limitations (6)
  • Test power falls when sequences differ little, so small or slowly evolving partitions may yield false negatives; substitution count accounts for under a fifth of pass/fail variance. · stated by the authors
  • The binomial extension (BiSymTest) ignores dependence among pairwise p-values caused by shared tree branches, so its results are relegated to the supplement. · stated by the authors
  • The three symmetry tests do not map one-to-one onto stationarity, reversibility and homogeneity, so which specific SRH condition is violated cannot be pinpointed. · stated by the authors
  • In the metazoan case study, the authors found no clear GC-content evidence explaining why failing partitions regroup clades, leaving the mechanism behind the topology shift unresolved. · stated by the authors
  • No simulations with known true trees are run, so the study shows that passing and failing partitions disagree but not that Tpass is more accurate; the accuracy benefit of exclusion remains untested. · noted in this summary
  • The abstract cites 39.5% of partitions rejecting SRH, whereas the results report 40.8% failing MaxSymTest and 50.5% failing any test, so it is unclear how the headline figure was derived. · noted in this summary

Read this if you build concatenated phylogenomic trees and want a permutation-based screen for non-stationary or non-homogeneous partitions, plus evidence of how such partitions shift topologies.

Citations
187
FWCI
6.45
Field percentile
top 2%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISNaser-Khdou '21Darriba '15Fabreti '22Gerth '19Duchene '18Wong '24
79% semantic matchEvolutionary Biology
S. Naser-Khdour, R. Lanfear, B. Q. Minh·bioRxiv·Sep 22, 2021