RAxML is run by a pre-created XML file in the CIPRES Portal:
The CIPRES portal currently runs RAxML on DNA and Protein sequences
Here is the command file:
<?xml version="1.0" encoding="UTF-8"?>
<command-object>
<command-target>
<reference>
<registry-info>cipres0,IDL:CipresIDL_api1/TreeImprove:1.0,raxml,Python wrapper around RAxMLHPC. file-based io.</registry-info>
</reference>
<ui-id>raxml</ui-id>
</command-target>
<commands>-c 25</commands>
<commands>-m GTRGAMMA</commands>
</command-object>
It sets only two flags:
-c number of categories; =25; and -m model Of Evolution; =GTRGAMMA
An explanation of the flags is below (the manual [pdf] provides more details):
-c Number Of Categories
This option allows you to specify the number of distinct rate categories used into which the individually optimized rates for each individual site are “thrown” under -m GTRCAT. The results in [Stamatakis, A.: Phylogenetic models of rate heterogeneity: A high performance computing perspective. In: Proc. of IPDPS2006, Rhodos, Greece (2006)] [pdf] indicate that the default of -c 25 works fine in most practical cases.
-m model Of Evolution i.e. Selection of the model of nucleotide substitution or amino acid substitution to be used.
NUCLEOTIDE MODELS (allowed values; CIPRES default is in bold)
-m GTRCAT
GTR approximation with optimization of individual per– site substitution rates and classification of those individual rates into the number of rate categories specified by -c. This is only a work-around for GTRGAMMA so make sure not to compare alternative topologies based on their GTRCAT likelihood values. Therefore, you can not use GTRCAT in combination with -f e (tree evaluation) and not in combination with multiple analyses on the original alignment (-#) option. This is due to the fact that the author assumes that you want to compare trees based on likelihoods if you do a multiple run on the original alignment. If you specify e.g. -m GTRCAT and -# 10 the program will automatically use GTRMIX (see below).
-m GTRMIX
This option will make RAxML perform a tree inference (search for a good topology) under GTRCAT. When the analysis is finished RAxML will switch its model to GTRGAMMA and evaluate the final tree topology under GTRGAMMA such that it yields stable likelihood values.
-m GTRGAMMA
GTR model of nucleotide substitution with the G model of rate heterogeneity. All model parameters are estimated by RAxML. The GTRGAMMA implementation uses 4 discrete rate categories which represents an acceptable trade-off between speed and accuracy. Note that, this has been hard-coded for performance reasons, i.e. the number of discrete rate categories can not be changed by the user.
