Computational protocol: Disproportionate Cochlear Length in Genus Homo Shows a High Phylogenetic Signal during Apes’ Hearing Evolution

Similar protocols

Protocol publication

[…] The phylogenetic signal measures the statistical dependence among observations for species related by a phylogenetic tree. The basic principle is to test whether a given tree better fits a set of species data observed at its tips as compared with the fit obtained when the same data have been randomly permuted across the tips (i.e., when the topology of the tree is destroyed). Our test was implemented via a phylogenetic generalised least squares (PGLS) approach. In PGLS mode, the phylogenetic tree is converted into a variance-covariance matrix, with the diagonal elements reporting the path length for each species (the root-to-tips distances; the variance) and the off-diagonal elements reporting the time of shared evolution for each pair of species (the distances from the root to the most recent common ancestor of each pair of species; the covariance). The covariance between the values in two tips of the tree is defined as the product of the trait values for the two tips, each measured as deviations from the ancestral state at the root node of the phylogeny. When two tips share a greater proportion of common history, their expected phylogenetic covariance is relatively high.In order to compute the phylogenetic covariances, we used the most common model for the evolution of continuously valued traits: the Brownian model []. Under this model, the expected variance for the trait value at a given tip of the tree is directly proportional to the summed branch length from the root to that tip. Therefore, the expected covariance between two values at the tips of the tree is directly proportional to the shared history of the taxa represented by the two tips.We used the likelihood ratio test to determine whether a Brownian model fitted our data and to compare two models of evolution: (i) a model that correspond to a standard Brownian constant-variance random-walk model with one parameter (variance of evolution) and (ii) a directional random-walk model with two parameters (variance of evolution and a parameter that reflects the degree of directional change). The likelihood ratio test compares the log-likelihood of the null hypothesis model (no directional trend exists) to that of the alternative hypothesis model (a directional trend exists) ().To measure the phylogenetic signal, we used the parameter lambda (λ) (Pagels’ λ) () which is multiplied to each off-diagonal elements in the variance-covariance matrix of shared evolutionary time between any pair of species in the gene-based phylogeny. The λ parameter reveals whether the phylogeny correctly predicts the patterns of covariance among species on a given trait, and its value can differ for different traits on the same phylogeny. The Pagels’ λ statistics varies between λ = 0 (the tree becomes more "star" like with all the branches emanating from a common node) and λ = 1 (the original tree is recovered). The λ parameter is typically estimated to obtain a value that maximizes the likelihood of the data. The statistical tests for a phylogenetic signal are successively performed under the null hypotheses that λ = 0, and that λ = 1. We reported tests for significant departure of λ from 0 and 1 (). The procedure is as follows: (i) we estimate the maximum likelihood (ML) value of λ in our data, and get the log- likelihood of this model; (ii) we run a model with λ fixed at its maximum value of 1; (iii) we use a likelihood ratio test to decide whether a model with ML λ fits the data better than a model with λ = 1. This tells us whether the phylogenetic signal in the data is equal or less than expected under the Brownian model given the phylogeny; (iv) we repeat the procedure and compare the model with ML λ with one in which λ = 0. A likelihood ratio test of a model with ML λ versus a model with λ = 0 will tell us if the phylogenetic signal in the data is greater than 0.The likelihood ratio test is calculated as: 2 * (log-likelihood of best fitting model—log-likelihood of worst fitting model). The best fitting model has the highest likelihood. The likelihood ratio is the absolute (i.e. positive) value of the difference between Log-likelihoods of the two competing nested models. We assess the significance of this value against a χ2 distribution with degrees of freedom (df) equal the difference in the number of estimated parameters between competing models (the directional random-walk model has two parameters and the Brownian motion model has one parameter). If the result is significant, then the directional random-walk model describes the data significantly better than the Brownian model, and should therefore be preferred, for instance when estimating ancestral states.More details can be found in Blomberg et al. [], Revell et al. [] and []. All the tests were made for each trait (cochlear parameters and body mass) separately on: (i) the entire sample of 22 catarrhines available in this study, (ii) the hominoid clade only (9 species), (iii) the cercopithecoid clade only (13 species). Because these calculations can be difficult when made using small numbers of species (i.e., less than 20), we consider our results based on all catarrhine species as more robust. We used the Unix executable program BayesTraits, Version 1 [] ( […]

Pipeline specifications

Software tools PHYSIG, BayesTraits
Application Phylogenetics
Organisms Homo sapiens, Gorilla gorilla
Diseases Foramen Ovale, Patent