New methods to calculate concordance factors for phylogenomic datasets

Key results

1. ShowngCF and sCF are both built into IQ-TREE 2: gCF counts only decisive gene trees (holding at least one taxon from each of the four clades around a branch), and sCF averages parsimony-informative site support over random quartets, each with matching discordance factors.

   New --gcf and --scf options can run together, and -S infers one tree per partition or alignment; gCF+gDF1+gDF2+gDFP and sCF+sDF1+sDF2 each sum to 1; a '.cf.stat' table holds all 16 values per branch, --cf-verbose adds per-gene and per-locus detail, and rooted reference plus gene trees let gCF differ across the root. · Gene Concordance Factor (gCF); Site Concordance Factor (sCF); gCF and sCF for rooted trees; Implementation in IQ-TREE 2

2. ShownNear-universal 100% bootstrap support on the lizard tree coexisted with low gene- and site-level concordance on some branches, showing that maximal bootstrap values do not guarantee high concordance.

   Lizard tree (43 species, 3,220 ultra-conserved elements, UCEs): all branches but one (Lipinia rouxi + L. pulchella, 62%) had 100% bootstrap, yet the minimum gCF and sCF were 4.5% and 33.2%; the same decoupling recurred in all 200 analyses across the 10 empirical datasets. · Application to empirical datasets; Interpretation of gCF and sCF values; Figure legends

3. ShownAdding loci tends to raise bootstrap values and the fraction of branches at 100%, whereas gCF and sCF vary mainly at small sample sizes and then stay nearly flat, consistent with bootstraps tracking sampling error and concordance tracking underlying variance.

   Each dataset was reanalysed 20 times (10 to 200 loci, in steps of 10) on the 200-locus concatenated maximum likelihood (ML) tree; bootstraps tended to rise in all 10 datasets, while concordance factors showed estimation error mainly in small subsamples (e.g., 30 or fewer lizard loci); standard and ultrafast bootstraps were very similar for lizards. · Application to empirical datasets; Interpretation of gCF and sCF values; Figure legends

4. ShownsCF rarely falls below about 33% because each decisive site supports one of three quartet topologies, whereas gCF can approach 0% when gene trees break monophyly of neighbouring clades; for lizards sCF was always at least gCF, which the authors attribute to short UCE loci.

   Empirical gCF ranged from 0% to 100% while sCF seldom fell under 33%; random alignments or saturated long branches should give the three quartet trees near-equal site support (sCF near 33%); gCF can also lose weight to the paraphyly class gDFP, which has no site-level analogue. · Site Concordance Factor (sCF); Interpretation of gCF and sCF values; Figure legends

5. ClaimedBranches with sCF under 33% mark splits that maximum parsimony (MP) would resolve differently from the reference tree, and are proposed as flags for reference trees misled by incomplete lineage sorting or driven by a few highly influential sites.

   Derived from the quartet definition of sCF; the authors point to parameter space where high incomplete lineage sorting misleads concatenated ML but not MP, and to outlier sites with extreme likelihood differences that MP weights like any other; no sub-33% branch is examined in the text. · Interpretation of gCF and sCF values