10 Jul 2013 Version 2.1.4 Revision 20130710

  * Added phyml auto-logging.
  * Added phyml command lines for best-fit models.
  * Added phyml log tab in the GUI.
  * Removed sample size modes (and "-n" argument). Sample size is fixed to alignment size.
  * Fixed bug with relative paths when calling from a different path.
  * Fixed typos in the GUI.

06 Mar 2013 Version 2.1.3 Revision 20130306

  * Fixed bug with PAUP* command block.
  * Added the possibility to change Inforation Criterion used with the clustering algorithm for the 203 matrices.
  * Changed "-o" argument for the hypothesis order into "-O".
  * Added "-o" argument for forwarding the standard output to a file: -o FILENAME

01 Jan 2013 Version 2.1.2 Revision 20130103

  * Fixed bug in paths with whitespaces.
  * Updated PhyML binaries. 

31 Jul 2012 Version 2.1.1 Revision 20120731

  * Fixed bug with hLRT selection when attempting to use a user-defined topology. 

11 Mar 2012 Version 2.1 Revision 20120511

Major updates:

  * Exhaustive GTR submodels: All the 203 different partitions of the GTR rate matrix can be included in the candidate set of models. When combined with rate variation (+I,+G, +I+G) and equal/unequal base frequencies the total number of possible models is 203 x 8 = 1624. 
  * Hill climbing hierarchical clustering: Calculating the likelihood score for a large number of models can be extremely time-consuming. This hill-climbing algorithm implements a hierarchical clustering to search for the best-fit models within the full set of 1624 models, but optimizing at most 288 models while maintaining model selection accuracy. 
  * Heuristic filtering: Heuristic reduction of the candidate models set based on a similarity filtering threshold among the GTR rates and the estimates of among-site rate variation. 
  * Absolute model fit: Information criterion distances can be calculated for the best-fit model against the unconstrained multinomial model (based on site pattern frequencies). This is computed by default when the alignment does not contain missing data/ambiguities, but can also be approximated otherwise. 
  * Topological summary: Tree topologies supported by the different candidate models are summarized in the html log, including confidence intervals constructed from cumulative models weights, plus Robinson-Foulds and Euclidean distances to the best-fit tree for each. 

Minor updates:

  * Corrected a bug in the fixed BIONJ-JC starting topology. F81+I+G was executed instead of JC.
  * "Best" is now the default tree search operation instead of NNI. "Best" computes both NNI and SPR algorithms and selects the best of them.
  * User can select the number of threads from GUI. 

1 Feb 2012 Version 2.0.2 Revision 20120201

  * Added a selection summary at the end of the console output.
  * Corrected the table header in the DT results frame (sorting).
  * Corrected a bug in DT Criterion where selection could not take place with large alignments.
  * Corrected a bug with command console version, where the execution crashed with certain arguments.
  * Unified LOCALE for English format. 

2 Nov 2011 Version 2.0.1

  * Improved thread scheduling algorithm.
  * OpenMP phyml patch for hybrid execution.
  * New argument (machinesfile) for hybrid execution on heterogeneous architectures, or heterogeneous resources distribution. 

Revision 20111013

  * Added conf/jmodeltest.conf file, where you can:
        Enable/Disable the automatic logging:

            You might be running a huge dataset and you don't want to generate hundreds or thousands of log files. 

        Set the PhyML binaries location:

            If you already have installed PhyML in your machine, you can setup jModelTest for use your own binaries. 

    Enhanced the html log output. 

27 Sep 2011

Version 2.0 Revision 20110927

  * High Performance Computing: The most important difference between jModelTest2 and the previous versions is the High Performance Computing (HPC) implementation. jModelTest2 is now capable to efficiently exploit current multicore desktop computers, while HPC clusters provide jModelTest2 with the highest speedup. The Message-Passing parallel implementation scales, in most common cases, up to half the total number of the candidate models. The total execution time can be improved even further to a lesser extent, up to the total number of candidate models.
  * Phylogenetic Averaging: jModelTest2 does not depend on external applications for phylogenetic averaging anymore. The native implementation allows now more flexibility for displaying intermediate results, as split support. Also different options can be taken into account, like the consensus threshold or the computation of consensus branch lengths (weighted median or weighted average).
  * Multiformat Input Data: jModelTest2 uses ALTER API for MSA format conversion, supporting now ALN, FASTA, GDE, MSF, NEXUS, PHYLIP and PIR formats.  
