Fits a Bunch of Models with PhyML
This function calls PhyML and fits successively 28 models of DNA evolution. The results are saved on disk, as PhyML usually does, and returned in R as a vector with the log-likelihood value of each model.
phymltest(seqfile, format = "interleaved", itree = NULL, exclude = NULL, execname = NULL, append = TRUE) ## S3 method for class 'phymltest': print(x, ...) ## S3 method for class 'phymltest': summary(object, ...) ## S3 method for class 'phymltest': plot(x, main = NULL, col = "blue", ...)
- a character string giving the name of the file that contains the DNA sequences to be analysed by PhyML.
- a character string specifying the format of the DNA
"interleaved"(the default), or
- a character string giving the name of a file with a tree
in Newick format to be used as an initial tree by PhyML. If
NULL(the default), PhyML uses a ``BIONJ'' tree.
- a vector of mode character giving the models to be excluded from the analysis. These must be among those below, and follow the same syntax.
- a character string specifying the name of the PhyML
executable. This argument can be left as
NULLif PhyML's default names are used:
- a logical indicating whether to erase previous PhyML output files if present; the default is to not erase.
- an object of class
- an object of class
- a title for the plot; if left
NULL, a title is made with the name of the object (use
main = ""to have no title).
- a colour used for the segments showing the AIC values (blue by default).
- further arguments passed to or from other methods.
The present function requires version 3.0.1 of PhyML; it won't work with older versions.
The user must take care to set correctly the three different paths involved here: the path to PhyML's binary, the path to the sequence file, and the path to R's working directory. The function should work if all three paths are different. Obviously, there should be no problem if they are all the same.
The following syntax is used for the models:
where "X" is the first letter of the author of the model, "Y" and "Z" are possibly other co-authors of the model, "00" is the year of the publication of the model, and "+I" and "+G" indicates whether the presence of invariant sites and/or a gamma distribution of substitution rates have been specified. Thus, Kimura's model is denoted "K80" and not "K2P". The exception to this rule is the general time-reversible model which is simple denoted "GTR" model.
The seven substitution models used are: "JC69", "K80", "F81", "F84",
"HKY85", "TN93", and "GTR". These models are then altered by adding
the "+I" and/or "+G", resulting thus in four variants for each of them
(e.g., "JC69", "JC69+I", "JC69+G", "JC69+I+G"). Some of these models
are described in the help page of
When a gamma distribution of substitution rates is specified, four categories are used (which is PhyML's default behaviour), and the ``alpha'' parameter is estimated from the data.
For the models with a different substition rate for transitions and transversions, these rates are left free and estimated from the data (and not constrained with a ratio of 4 as in PhyML's default).
path2exec has been removed in the present version:
the path to PhyML's executable can be specified with the option
phymltestreturns an object of class
"phymltest": a numeric vector with the models as names.
"phymltest"as matrix with the name of the models, the number of free parameters, the log-likelihood value, and the value of the Akaike information criterion (AIC = -2 * loglik + 2 * number of free parameters)
summarymethod prints all the possible likelihood ratio tests for an object of class
plotmethod plots the values of AIC of an object of class
"phymltest"on a vertical scale.
It is important to note that the models fitted by this function is only a small fraction of the models possible with PhyML. For instance, it is possible to vary the number of categories in the (discretized) gamma distribution of substitution rates, and many parameters can be fixed by the user. The results from the present function should rather be taken as indicative of a best model.
Posada, D. and Crandall, K. A. (2001) Selecting the best-fit model of nucleotide substitution. Systematic Biology, 50, 580--601.
Guindon, S. and Gascuel, O. (2003) A simple, fast, and accurate
algorithm to estimate large phylogenies by maximum likelihood.
Systematic Biology, 52, 696--704.
### A `fake' example with random likelihood values: it does not ### make sense, but does not need PhyML and gives you a flavour ### of what the output looks like: x <- runif(28, -100, -50) names(x) <- .phymltest.model class(x) <- "phymltest" x summary(x) plot(x) plot(x, main = "", col = "red") ### This example needs PhyML, copy/paste or type the ### following commands if you want to try them, eventually ### changing setwd() and the options of phymltest() setwd("D:/phyml_v2.4/exe") # under Windows data(woodmouse) write.dna(woodmouse, "woodmouse.txt") X <- phymltest("woodmouse.txt") X summary(X) plot(X)