Exploring the impact of unstable terminals on branch support values in paleontological data

Recent studies have acknowledged the many benefits of including fossils in phylogenetic inference (e.g., reducing long-branch attraction). However, unstable taxa are known to be problematic, as they can reduce either the resolution of the strict consensus or branch support. In this study, we evaluate whether unstable taxa that reduce consensus resolution affect support values, and the extent of such impact, under equal and extended implied weighting. Two sets of analyses were conducted across 30 morphological datasets to evaluate complementary aspects. The first focused on the analytical conditions incrementing the terminal instability, while the second assessed whether pruning wildcards improves support. Changes in support were compared with the “number of nodes collapsed by unstable terminals,” their “distance to the root,” the “proportion of missing data in a dataset,” and the “proportion of sampled characters.” Our results indicate that the proportion of missing entries distributed among closely related taxa (for a given character) might be as detrimental for stability as those distributed among characters (for a given terminal). Unstable terminals that (1) collapse few nodes or (2) are closely located to the root node have more influence on the estimated support values. Weighting characters according to their extra steps while assuming that missing entries contribute to their homoplasy reduced the instability of wildcards. Our results suggest that increasing character sampling and using extended implied weighting decreases the impact of wildcard terminals. This study provides insights for designing future research dealing with unstable terminals, a typical problem of paleontological data.

Taxonomic sampling and rare genomic changes overcome long-branch attraction in the phylogenetic placement of pseudoscorpions

Long-branch attraction is a systematic artifact that results in erroneous groupings of fast-evolving taxa. The combination of short, deep internodes in tandem with LBA artifacts has produced empirically intractable parts of the Tree of Life. One such group is the arthropod subphylum Chelicerata, whose backbone phylogeny has remained unstable despite improvements in phylogenetic methods and genome-scale datasets. Pseudoscorpion placement is particularly variable across datasets and analytical frameworks, with this group either clustering with other long-branch orders or with Arachnopulmonata (scorpions and tetrapulmonates). To surmount LBA, we investigated the effect of taxonomic sampling via sequential deletion of basally branching pseudoscorpion superfamilies, as well as varying gene occupancy thresholds in supermatrices. We show that concatenated supermatrices and coalescent-based summary species tree approaches support a sister group relationship of pseudoscorpions and scorpions, when more of the basally branching taxa are sampled. Matrix completeness had demonstrably less influence on tree topology. As an external arbiter of phylogenetic placement, we leveraged the recent discovery of an ancient genome duplication in the common ancestor of Arachnopulmonata as a litmus test for competing hypotheses of pseudoscorpion relationships. We generated a high-quality developmental transcriptome and the first genome for pseudoscorpions to assess the incidence of arachnopulmonate-specific duplications (e.g., homeobox genes and miRNAs). Our results support the inclusion of pseudoscorpions in Arachnopulmonata (new definition), as the sister group of scorpions. Panscorpiones (new name) is proposed for the clade uniting Scorpiones and Pseudoscorpiones.

Taxonomic sampling and rare genomic changes overcome long-branch attraction in the phylogenetic placement of pseudoscorpions

Long-branch attraction is a systematic artifact that results in erroneous groupings of fast-evolving taxa. The combination of short, deep internodes in tandem with LBA artifacts has produced empirically intractable parts of the Tree of Life. One such group is the arthropod subphylum Chelicerata, whose backbone phylogeny has remained unstable despite improvements in phylogenetic methods and genome-scale datasets. Pseudoscorpion placement is particularly variable across datasets and analytical frameworks, with this group either clustering with other long-branch orders or with Arachnopulmonata (scorpions and tetrapulmonates). To surmount LBA, we investigated the effect of taxonomic sampling via sequential deletion of basally branching pseudoscorpion superfamilies, as well as varying gene occupancy thresholds in supermatrices. We show that concatenated supermatrices and coalescent-based summary species tree approaches support a sister group relationship of pseudoscorpions and scorpions, when more of the basally branching taxa are sampled. Matrix completeness had demonstrably less influence on tree topology. As an external arbiter of phylogenetic placement, we leveraged the recent discovery of an ancient genome duplication in the common ancestor of Arachnopulmonata as a litmus test for competing hypotheses of pseudoscorpion relationships. We generated a high-quality developmental transcriptome and the first genome for pseudoscorpions to assess the incidence of arachnopulmonate-specific duplications (e.g., homeobox genes and miRNAs). Our results support the inclusion of pseudoscorpions in Arachnopulmonata, as the sister group of scorpions. Panscorpiones (new name) is proposed for the clade uniting Scorpiones and Pseudoscorpiones.

Scalable Empirical Mixture Models That Account for Across-Site Compositional Heterogeneity

Biochemical demands constrain the range of amino acids acceptable at specific sites resulting in across-site compositional heterogeneity of the amino acid replacement process. Phylogenetic models that disregard this heterogeneity are prone to systematic errors, which can lead to severe long-branch attraction artifacts. State-of-the-art models accounting for across-site compositional heterogeneity include the CAT model, which is computationally expensive, and empirical distribution mixture models estimated via maximum likelihood (C10–C60 models). Here, we present a new, scalable method EDCluster for finding empirical distribution mixture models involving a simple cluster analysis. The cluster analysis utilizes specific coordinate transformations which allow the detection of specialized amino acid distributions either from curated databases or from the alignment at hand. We apply EDCluster to the HOGENOM and HSSP databases in order to provide universal distribution mixture (UDM) models comprising up to 4,096 components. Detailed analyses of the UDM models demonstrate the removal of various long-branch attraction artifacts and improved performance compared with the C10–C60 models. Ready-to-use implementations of the UDM models are provided for three established software packages (IQ-TREE, Phylobayes, and RevBayes).

Distinguishing Felsenstein Zone from Farris Zone Using Neural Networks

Maximum likelihood and maximum parsimony are two key methods for phylogenetic tree reconstruction. Under certain conditions, each of these two methods can perform more or less efficiently, resulting in unresolved or disputed phylogenies. We show that a neural network can distinguish between four-taxon alignments that were evolved under conditions susceptible to either long-branch attraction or long-branch repulsion. When likelihood and parsimony methods are discordant, the neural network can provide insight as to which tree reconstruction method is best suited to the alignment. When applied to the contentious case of Strepsiptera evolution, our method shows robust support for the current scientific view, that is, it places Strepsiptera with beetles, distant from flies.

Supermatrix analyses and molecular clock rooting of Fabales: Exploring the effects of outgroup choice and long branch attraction on topology

Fabales is a cosmopolitan angiosperm order that consists of four families: Leguminosae (Fabaceae), Polygalaceae, Surianaceae, and Quillajaceae. Despite the great interest in this group, a convincing phylogeny of the order is still not available. Therefore, the aim of this study was to explicitly test for possible long branch attraction (LBA) problems within Fabales for the first time, and determine whether low tree stemminess and unequal branch lengths could worsen this problem. Supermatrix analysis of Fabales was carried out using previously published plastid matK, trnL, rbcL, and newly sequenced nuclear sqd1 regions for 678 taxa in total, including 43 outgroup taxa from families of Fabidae. We employed additional analyses, such as simulations, network analyses, sampling different outgroup taxa (random or real), removing fast evolving sites and fast evolving taxa, and molecular clock rooting, to identify both LBA and (or) rooting problems. These analyses clearly show that the Fabales phylogeny has been influenced by the sampling of outgroup taxa, but not LBA. However, network analyses show that even though it is weak, there is a consistent phylogenetic signal among the rapidly radiated Fabales families, which can be traced by further analyses. While, molecular clock rooting analysis yielded a (Leguminosae(Polygalaceae(Surianaceae+Quillajaceae))) topology with strong support for the first time here, supermatrix analyses yielded a ((Leguminosae+Polygalaceae)(Surianaceae+Quillajaceae)) with low-moderate support.

Architectural instability, inverted skews and mitochondrial phylogenomics of Isopoda: outgroup choice affects the long-branch attraction artefacts

The majority strand of mitochondrial genomes of crustaceans usually exhibits negative GC skews. Most isopods exhibit an inversed strand asymmetry, believed to be a consequence of an inversion of the replication origin (ROI). Recently, we proposed that an additional ROI event in the common ancestor of Cymothoidae and Corallanidae families resulted in a double-inverted skew (negative GC), and that taxa with homoplastic skews cluster together in phylogenetic analyses (long-branch attraction, LBA). Herein, we further explore these hypotheses, for which we sequenced the mitogenome of Asotana magnifica (Cymothoidae), and tested whether our conclusions were biased by poor taxon sampling and inclusion of outgroups. (1) The new mitogenome also exhibits a double-inverted skew, which supports the hypothesis of an additional ROI event in the common ancestor of Cymothoidae and Corallanidae families. (2) It exhibits a unique gene order, which corroborates that isopods possess exceptionally destabilized mitogenomic architecture. (3) Improved taxonomic sampling failed to resolve skew-driven phylogenetic artefacts. (4) The use of a single outgroup exacerbated the LBA, whereas both the use of a large number of outgroups and complete exclusion of outgroups ameliorated it.

Distinguishing Felsenstein zone from Farris zone using neural networks

Maximum likelihood and maximum parsimony are two key methods for phylogenetic tree reconstruction. Under certain conditions, each of these two methods can perform more or less efficiently than the other. We show that a neural network can efficiently distinguish between four-taxon alignments that were evolved under conditions conducive to long-branch attraction, or long-branch repulsion. The feedback from the neural network can be used to select the most efficient tree reconstruction method yielding increased accuracy, when compared to a rigid choice of reconstruction methods. When applied to the contentious case of Strepsiptera evolution, our method agrees with the current scientific view.

Scalable empirical mixture models that account for across-site compositional heterogeneity

Biochemical demands constrain the range of amino acids acceptable at specific sites resulting in across-site compositional heterogeneity of the amino acid replacement process. Phylogenetic models that disregard this heterogeneity are prone to systematic errors, which can lead to severe long branch attraction artifacts. State-of-the-art models accounting for across-site compositional heterogeneity include the CAT model, which is computationally expensive, and empirical distribution mixture models estimated via maximum likelihood (C10 to C60 models). Here, we present a new, scalable method EDCluster for finding empirical distribution mixture models involving a simple cluster analysis. The cluster analysis utilizes specific coordinate transformations which allow the detection of specialized amino acid distributions either from curated databases, or from the alignment at hand. We apply EDCluster to the HOGENOM and HSSP databases in order to provide universal distribution mixture (UDM) models comprising up to 4096 components. Detailed analyses of the UDM models demonstrate the removal of various long branch attraction artifacts and improved performance compared to the C10 to C60 models. Ready-to-use implementations of the UDM models are provided for three established software packages (IQ-TREE, Phylobayes, and RevBayes).