Fig. 1

Results of posterior predictive analyses of phylogenomic datasets of Shih et al. [10], Ponce-Toledo et al. [12], and Ochoa de Alda et al. [14]. Rows within each panel correspond to phylogenomic datasets. Observed values calculated for each test statistic are represented by vertical lines. Color and patterns of vertical lines indicate amino acid recoding strategies, respectively NR: No recoding, DAY6: six-state Dayhoff recoding, SR6: the six-state recoding strategy of Susko and Roger [59], and KGB6: the six-state recoding strategy of Kosio et al. [60]. Symbols show average values for two posterior predictive test statistics calculated from simulated datasets, with error bars showing ± five standard deviations. Symbol shapes correspond to phylogenetic models (LG: LG+4 Γ, CAT: CAT-GTR+4 Γ) and symbol colors show recoding strategies. If similarly colored error bars contain vertical lines, the given phylogenetic model adequately describes systematic biases of the given data. a Results with the PPA-VAR statistic [22] assessing fit of models to site-process heterogeneity in the data. b Results with the PPA-MAX statistic, measured as squared standard deviation (SSD) of amino acid composition [23], showing generally poor fit of models against the lineage-specific compositional heterogeneity in the data