The impact of incorporating molecular evolutionary model into predictions of phylogenetic signal and noise

Abstract

Phylogenetic inference can be improved by the development and use of better models for inference given the data available, or by gathering more appropriate data given the potential inferences to be made. Numerous studies have demonstrated the crucial importance of selecting a best-fit model to conducting accurate phylogenetic inference given a data set, explicitly revealing how model choice affects the results of phylogenetic inferences. However, the importance of specifying a correct model of evolution for predictions of the best data to be gathered has never been examined. Here, we extend analyses of phylogenetic signal and noise that predict the potential to resolve nodes in a phylogeny to incorporate all time-reversible Markov models of nucleotide substitution. Extending previous results on the canonical four-taxon tree, our theory yields an analytical method that uses estimates of the rates of evolution and the model of molecular evolution to predict the distribution of signal, noise, and polytomy. We applied our methods to a study of 29 taxa of the yeast genus Candida and allied members to predict the power of five markers, COX2, ACT1, RPB1, RPB2, and D1/D2 LSU, to resolve a poorly supported backbone node corresponding to a clade of haploid Candida species, as well as 19 other nodes that are reasonably short and at least moderately deep in the consensus tree. The use of simple, unrealistic models that did not take into account transition/transversion rate differences led to some discrepancies in predictions, but overall our results demonstrate that predictions of signal and noise in phylogenetics are fairly robust to model specification.

Tales from the crypt: genome mining from fungarium specimens improves resolution of the mushroom tree of life

“…In the present study, we used five metrics that maximize data quantity (alignment length), attempt to model signal directly (PHYDESIGN; Townsend et al ., ) or use bipartition frequency (mean bootstraps) as a proxy for signal, or minimize topological incongruence amongst the individual gene trees (Robinson–Foulds distances, TC; Salichos & Rokas, ). Studies that have used signal and noise estimation in PHYDESIGN as a guide for gene selection have shown that they perform well in resolving nodes at diverse time scales and taxonomic groups (López‐Giráldez, Moeller & Townsend, ; Su et al ., ). Alternatively, explicit attempts to reduce gene tree incongruence using measures of mean bootstraps or minimizing topological discordance have also shown improvement in phylogeny estimation (Townsend et al ., ).…”

Utility of characters evolving at diverse rates of evolution to resolve quartet trees with unequal branch lengths: analytical predictions of long-branch effects

“…We build on the signal and noise framework of Townsend et al [ 56 ], which uses the estimated substitution rates of individual molecular characters to estimate the power of a set of molecular sequences for resolving a four-taxon tree with equal subtending branch lengths. This result, applied to the Poisson model of molecular evolution, was subsequently generalized by Su et al [ 57 ] to apply to all standard symmetric molecular evolutionary models of nucleotide substitution up to and including the General Time Reversible model (GTR [ 58 , 59 ]). Herein we further generalize the signal and noise analysis by relaxing the assumption of equal subtending branch lengths for the four-taxon tree.…”

“…The Markov chain of a nucleotide character under the GTR model is commonly mathematically modeled by a four-by-four substitution rate matrix Q ( λ ), whose element q ij gives the instantaneous rate at which the nucleotide character changes from nucleotide i to nucleotide j , where j ≠ i, and i, j = T, C, A, or G ( c.f. Equation 1 in [ 57 ]). The average substitution rate of the character, λ , can be calculated as …”

“…Equation 2 can be solved via eigendecomposition ( c.f. [ 57 ]). Using P ( λ , t ), we track the Markov chain of a nucleotide character in an ultrametric four-taxon tree with four uneven subtending branches.…”

“…C 1 = C 3 ≠ C 2 = C 4 , or C 1 = C 4 ≠ C 2 = C 3 in Figure 1 ) contributes to incorrect resolution of the tree. Summing the probabilities of all possible scenarios of character state changes across the internode and subtending branches that result in a desired pattern of character states at the four terminal tips as in Su et al [ 57 ], the probability of a nucleotide character showing the synapomorphic pattern is provided by …”