Phylogenetic analysis of molecular sequence data using the maximum-likelihood criterion
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


// GARLI Version 2.1 (September 2014)
// Copyright 2005-2014 Derrick J. Zwickl
// email:
//  This program is free software: you can redistribute it and/or modify
//  it under the terms of the GNU General Public License as published by
//  the Free Software Foundation, either version 3 of the License, or
//  (at your option) any later version.
//  This program is distributed in the hope that it will be useful,
//  but WITHOUT ANY WARRANTY; without even the implied warranty of
//  GNU General Public License for more details.
//  You should have received a copy of the GNU General Public License
//  along with this program.  If not, see <>.

Please let me know of any problems, concerns or feedback (

GARLI version 2.1 is a minor update to version 2.0.
Versions >= 2.0 include models for nucleotides, amino acids,
odons, and morphology-like characters, any of which can be 
mixed together and applied to different subsets of data.

Version 2.X should replace earlier versions, and should be backwards compatible with all 
configuration files and datasets that were used with the previous versions.

->See the support website ( for detailed information on
using the program.  For very basic usage see QuickStart.txt.

->Example datasets and template configuration files files can be found in the example folder

->For compilation help see the INSTALL file.  Versions >= 2.0 comes with an extremely easy build
script,, which should make compalation trivial on linux or OS X machines.

***New in version 2.1***
1. MUCH faster parsing of very long alignments.  e.g., alignments millions of nucleotides in 
   length should be read thousands of times faster.
2. Lower memory usage with very large alignments.
3. Added ignorestopcodons entry to ignore stops rather than error out.
4. Better translation to amino acid characters in codon-aminoacid models. Codons containing ambiguity 
   are no longer discarded if all resolutions of the ambiguity translate to the same amino acid.
5. Ability to read and use NEXUS wtsets to specify character counts.
6. Many changes in BOINC mode.
7. Fixes to minor/rare bugs.

***New in version 2.0***

1. Ability to use partitioned models, giving the ability to divide up data and apply independent
   models to each.   See this page for details on partitioned usage
2. Ability to use the Mk/Mkv "morphology" models of Lewis, 2001.  This can be applied to discrete
   data of any type with any number of states.
3. Significant improvement in parameter optimization
4. MANY minor improvements and new features.  See the website.

***New in version 1.0***

1. Ability to write sitewise log-likelihood values for all model types, in a format identical to 
   PAUP*.  This can be read directly into a program like CONSEL to perform statistical comparisons 
   of topologies such as the SH or AU tests. (outputsitelikelihoods = 1)
2. Ability to collapse zero length branches at the end of any search, creating polytomies. This is
   now turned on by default. This setting can affect bootstrap values, since zero-length branches in
   a sense don't exist and probably shouldn't contribute to branch support values. It also tends to
   make trees inferred from multiple searches more similar, since trees that only differ in arbitrary
   resolutions of zero length branches are not truly different.  (collapsebranches = 1)
3. Ability to infer full reversible amino acid rate matrices while doing a normal searching,
   adding 189 free parameters.  This is probably not something of general utility unless you have
   a very large dataset. (datatype = aminoacid or codon-aminoacid, ratematrix = estimate) 
4. Ability to use user-specified amino acid rate matrices. This allows the use of any existing
   amino acid matrix, regardless of whether GARLI implements them internally.  Amino acid matrices 
   estimated by GARLI can also have their parameter values fixed for use in other analyses. Note 
   however that GARLI's matrix input format differs from other programs. (ratematrix = fixed, 
   provide matrix in a Nexus GARLI block in the datafile or in a starting conditions file.  See the
   file "examples/LGmodel.mod" for an example.)
5. Ability to infer internal state probabilities (ancestral states) for amino acid and codon models,
   in addition to the previously implemented nucleotide models. (inferinternalstateprobs = 1)
6. Substantial speed improvements for large constrained searches, especially backbone constraints
7. MPI parallel runs can now be checkpointed, allowing entire sets of runs to be restarted.  Be sure
   to read the wiki page detailing the MPI version (
   to understand in what cases you might want to use this version.
8. More rigorous error checking of input trees, constraints and parameter values. 
9. Significant improvements to the precision of parameter optimization.  GARLI now puts 
   significant effort into returning the very most optimal parameter values at the end of a search.
   These should be as accurate as values returned by other programs such as PAUP* or PAML.  
   Previously the estimated parameter values were nearly optimal, but sometimes not quite there.
10. A "verification mode", which checks that a given configuration file and datafile are valid
    for use with GARLI, without starting an actual analysis.  This can be useful, for example, in
    verifying that all configuration and input is proper while on your local machine before sending
    the input files to a computer cluster.  The output will also tell you how much memory GARLI
    will be need to be allocated for the run, which might require adjustment of the 
    "availablememory" setting in the configuration file (start GARLI with "-V").
11. Much easier procedure for compiling of GARLI source code.
12. Fixes to numerous rare bugs in version 0.96

***New in version 0.96***

1.Rigorous reading of Nexus datasets using Paul Lewis and Mark Holder's Nexus
 Class Library (NCL).
2.Ability to read Nexus starting trees using NCL.
3.Ability to perform inference under amino acid and codon-based models
 of sequence evolution (datatype = aminoacid, datatype = codon).
4.Ability to specify multiple search replicates in a single config file (searchreps = #).
5.Ability to specify outgroups for orientation of inferred trees (outgroup = # # #).
6.Ability to use backbone as well as normal topological constraints.
7.Ability to create fast likelihood stepwise addition starting trees
 (streefname = stepwise).
8.MPI version that spreads a specified number of serial runs across processors using
 a single config file, writing output to different output files (for example, to do
 25 bootstrap replicates simultaneously on each of 8 processors).
9.Ability to perform nucleotide inference using any sub-model of the General
 Time-Reversible model (GTR), in addition to all of the common "named" models (K2P, HKY, etc).
10.Speed increases for non-parametric bootstrapping

Condensed summary of new model settings (for more detailed descriptions and 
for unchanged settings see the manual or support webpage):
datatype = {nucleotide, aminoacid, codon, codon-aminoacid} - These set the type of model
 to be used.  Note that aminoacid and codon models are MUCH slower than nucleotide models.
-"aminoacid" is for datasets consisting of the 20 aminoacid single letter codes. 
-"codon" is for dna data (aligned in frame!) to be analyzed using a 60-62 state model
 that incorporates both the nucleotide substitution process and information on the
 genetic code.  This involves the estimation of at least one dN/dS ratio
 (aka nonsynonymous/synonymous rate ratio, or omega or w).  This is essentially
 the Goldman-Yang 1994 model and other related models.
-"codon-aminoacid" is for aligned dna sequences that are translated to aminoacids
 and analyzed under an aminoacid model

The different datatypes have different allowable model settings, listed here.
For nucleotide data:
ratematrix = {6rate, 2rate, 1rate, fixed, (a b c d e f) }
statefrequencies = {estimate, empirical, equal, fixed}
ratehetmodel = {none, gamma, gammafixed}
numratecategories = {#} (not including invariant site class, must be 1 for ratehetmodel = none)
invariantsites = {estimate, none}

For aminoacid or codon-aminoacid data:
ratematrix = {poisson, dayhoff, jones, wag, mtmam, mtrev}
statefequencies = {equal, dayhoff, jones, wag, mtmam, mtrev, empirical, estimate}
ratehetmodel = {none, gamma, gammafixed}
numratecategories = {#} (not including invariant site class, must be 1 for ratehetmodel = none)
invariantsites = {estimate, none, fixed}

For codon data:
ratematrix = {6rate, 2rate, 1rate, fixed, (a b c d e f) }
 (this is the nucleotide substitution process assumed by the codon model)
statefrequencies = {empirical, equal, F1x4, F3x4} (F1x4 and F3x4 are PAML's
 terminology, and calculate the codon frequencies as the product of the total
 nucleotide frequencies or the nucleotide frequencies at each codon position, respectively)
ratehetmodel = {none, nonsynonymous} {nonsynonymous estimates multiple dN/dS categories
 at the proportion of sites belonging to each.  This is the M3 model of PAML)
numratecategories = {#} (the number of dN/dS categories)
invariantsites = {none} (not allowed in codon model)

For codon or codon-aminoacid:
Geneticcode = {standard, vertmito, invertmito}