Systematic errors in phylogenetic trees

The effort to reconstruct the tree of life was revolutionized by use sequences proteins and nucleic acids. Phylogenetic trees are now routinely inferred using hundreds thousands amino acid or nucleotide characters. It thus seems surprising that many aspects still controversial; conflicting results between large scale phylogenomic studies show errors remain common despite datasets. These often result from systematic biases in way evolve. While resulting well understood, it requires careful efforts reduce their effects. fundamentals molecular phylogenetics straightforward: aligning homologous genes (those inherited a ancestor), individual nucleotides acids can be identified different species. heritable substitutions these sites gene protein experience lineages over evolutionary history passed on descendants constitute record species relationships. most source error phylogenetic reconstruction is homoplasy, whereby same novel character appears two through convergent evolution rather than because both have ancestor. Homoplasy inevitable sequence data due limited set states any given site an alignment adopt — four 20 types error: stochastic small samples (and mostly eliminated sets); more devious error. Systematic consistent repeatable faulty assumptions analysis. Such commonly arise when models we our assume process change homogeneous reality heterogeneous either across sites, taxa time. Ignoring heterogeneities sometimes artefacts inference, as added, certain see effects In this primer, will focus three best-known heterogeneity which, ignored, may inference errors: first, rates lineages; second, within alignment; third, state composition sites. We simple short synthetic alignments illustrate how ubiquitous features cause reconstructing trees. Evolutionary probability one another time period vary substantially species, especially distantly related taxa. Two unrelated, fast-evolving organisms each tend accumulate mutations. Given diversity acids, there risk evolve new at protein. Their slowly evolving relatives less likely convergence. If asymmetry outcomes not accounted for, misinterpreting homoplasy for true signal clustering unrelated together. phenomenon fast incorrectly grouped infamous long-branch-attraction (LBA)artefact. examples, just such phylogeny (Figure 1) involving relatively arthropod vertebrate rapidly nematode ctenophore (comb jelly) which outgroup other correct groups with arthropod. separated branch contains little evidence separation. variation expectation obtain long attraction topology long-branched attracted leading outgroup. corollary short-branched placed closest relative vertebrate. maximum parsimony tree-building method treats all equally, therefore does explicitly model potential difference Maximum chooses minimises number required explain observed data. Only patterns imply numbers changes topologies. Any characters ‘parsimony informative’ ignored method. example seven 1A), only first informative; last four, occurred along branches single taxon (importantly purposes, taxa). Each explained substitution topology, meaning they uninformative. Of informative characters, 1A, red triangle) supports share nucleotide, mutation contrast, next (indicated blue stars) (nematode ctenophore) evolved independent mutations branches. When compare possible topologies, five 1B). convergently (blue) must twice, plus character. By tree, interpreted having changed once separating ctenophores nematodes vertebrates arthropods. counted experiencing changes). distribution preferred parsimonious tree. likelihood-based methods (maximum likelihood Bayesian approaches) explicit estimate observing lengths. Branch lengths reflect expected per dataset, lot has rate high. To give example, very similar high being low branch. 2A), greater its ancestor branch, reflecting few accumulated. On contrary, nematode, longer higher likelihood. Knowing totality also feeds back expectations changing increases increasing length 2B). end if Accounting affects 2C). Using before, topologies non-informative included. Under likelihood, uninformative provide information about ctenophore. As add taxa, get longer. so becomes homoplastic twice. added reveal branches, Parsimony effectively underestimates hence Branch-length aware accommodate branch-length topology. Apart evolution, observe alignment. Perhaps obvious (or concatenated alignment). Individual domains whole under selective pressures rates. At level coding sequence, third position codon typically silent, do occur much frequently positions. Combined (unequal lengths), variance been shown fail recover failure differences (average) means systematically faster-evolving overestimates slower-evolving Because those undergo assuming artefacts. Consider positions quickly second 3). ignore uniform underestimate codon-position might convergently. prefers If, however, partition position, find that, faster 3 1 2, decreases, increases. partitions highest produced rates, general among-site variation, modelled random variable following gamma distribution. This strategy accounting among widely used. Different residues but restricted subsets according function residue. A clear would regions protein, crosses membrane extracellular; former constrained hydrophobic latter comprised predominantly hydrophilic Most compositional homogeneity For uses average leucine isoleucine measured real functional reasons, isoleucine. between-site 4). simplicity, consider DNA although probably important proteins. nucleotides, ACGT, found equal frequencies. extreme heterogeneity, categories frequencies nucleotides; partitions, contain GC AT respectively. estimated transition G C lower frequency (0.25 each) (0.5 each). underestimating (GC-rich) (AT-rich) partitions. leads and, result, composition, able correctly identify instances conditions, corresponds Our examples designed ignoring inference. partitioning into applying suitable them overcome errors. data, don’t necessarily know priori rely mixture multiple sets parameters likelihoods. Among Gamma model, accommodates across-site CAT (categories), C10–C60 models, variance. heterogeneously suppress otherwise result. described sources error, empirical revealed several deviations Composition AT- GC-rich genome. some constant (heterotachy) (heteropecilly). harder currently try remove suffering biases. choice increasingly inferring phylogenies involve heated debates concerning various nodes revolve around suspicion Famous include whether amitochondriate microsporidia members fungi; ecdysozoans, including arthropods, monophyletic; affinities xenacoelomorph worms. cases complex computational resources considerable. Without anticipate errors, cannot resolve difficult parts remaining life.