Testing version 1.4.0-beta3

NOTE: This is a beta version for testing purpose. Please send feedback if you think it does not produce the results as expected, e.g., when compared with IQ-TREE 1.3.X, TREE-PUZZLE or CONSEL.

Updates in beta3:

• Fix a numerical issue when computing maximum likelihood distances (reported by Giap Nguyen)

Updates in beta2:

• Fix a bug in likelihood mapping (reported by Giap Nguyen)
• Support partition models for likelihood mapping

New features:

• Checkpointing: If an IQ-TREE run was stopped for whatever reason, rerunning it with the very same command-line will automatically recover and continue the previous run. If a run successfully finished, IQ-TREE will deny to rerun and issue an error message. In such cases, one can start a new run using another prefix (-pre option) or tell IQ-TREE to overwrite all output files (-redo option).

• Implementation of the likelihood mapping approach (Strimmer and von Haeseler, 1997) to assess the phylogenetic information of an input alignment via -lmap option, which specifies the number of quartets to be drawn randomly. Compared with the original implementation of TREE-PUZZLE, this implementation supports many more substitution models available in IQ-TREE (incl. partition and mixture models). Example usage:

  iqtree -s example.phy -m GTR+G -n 0 -lmap 10000
  

  The likelihood mapping plot will be printed to .svg and .eps file. The report file .iqtree will contain a new section LIKELIHOOD MAPPING STATISTICS for the results.

• Implementation of the approximately unbiased (AU) test (Shimodaira, 2002) via -au option. Compared with the original implementation of CONSEL, this implementation is partition-aware (i.e., bootstrap resampling sites within partitions). Example usage:

  iqtree -s example.phy -m GTR+G -n 0 -z example.trees -zb 10000 -au
  

Testing version 1.4.0-beta2

NOTE: This is a beta version for testing purpose. Please send feedback if you think it does not produce the results as expected, e.g., when compared with IQ-TREE 1.3.X, TREE-PUZZLE or CONSEL.

Updates in beta2:

• Fix a bug in likelihood mapping (reported by Giap Nguyen)
• Support partition models for likelihood mapping

New features:

• Checkpointing: If an IQ-TREE run was stopped for whatever reason, rerunning it with the very same command-line will automatically recover and continue the previous run. If a run successfully finished, IQ-TREE will deny to rerun and issue an error message. In such cases, one can start a new run using another prefix (-pre option) or tell IQ-TREE to overwrite all output files (-redo option).

• Implementation of the likelihood mapping approach (Strimmer and von Haeseler, 1997) to assess the phylogenetic information of an input alignment via -lmap option, which specifies the number of quartets to be drawn randomly. Compared with the original implementation of TREE-PUZZLE, this implementation supports many more substitution models available in IQ-TREE (incl. partition and mixture models). Example usage:

  iqtree -s example.phy -m GTR+G -n 0 -lmap 10000
  

  The likelihood mapping plot will be printed to .svg and .eps file. The report file .iqtree will contain a new section LIKELIHOOD MAPPING STATISTICS for the results.

• Implementation of the approximately unbiased (AU) test (Shimodaira, 2002) via -au option. Compared with the original implementation of CONSEL, this implementation is partition-aware (i.e., bootstrap resampling sites within partitions). Example usage:

  iqtree -s example.phy -m GTR+G -n 0 -z example.trees -zb 10000 -au
  

Version 1.3.13

Bug fixes:

• Fix a numerical underflow with ascertainment bias correction [+ASC] model for large data sets (reported by Alex Riedel)

Version 1.3.12

Bug fixes:

• A crash when performing analysis on partitioning schemes produced by the k-means algorithm of PartitionFinder (reported by Pete Hosner). This is due to a partition containing only constant sites and the other partitions containing only variable sites. Such partitioning schemes are not recommended for phylogenetic analysis and users are advised to consult PartitionFinder author.
• A crash with too high or too low rates for FreeRate model (reported by Hermes Escalona and Adrian Arellano Davin).

IQ-TREE version 1.3.11

New features:

• For long alignments (>100,000 sites) the minimal branch length is now reduced to 0.1/alignment_length to accommodate analysis of very closely related sequences (thanks to David Wyllie for testing).
• New -blmin and -blmax option for min and max branch length (requested by Hang Phan).
• New -wslm and -wslmr options to print site log-likelihood per mixture class and per mixture+rate category, respectively (requested by Huaichun Wang).
• New --link-alpha option to link Gamma shape parameter (alpha) across partitions (requested by Huaichun Wang).

Bug fixes:

• Numerical problems with +ASC model for protein data: disabled now by default (reported by several web server users and Lars Jermiin).
• Fix a rare crash for option combination -z and -sp when changing to old kernel.
• Fix a crash with multicore version with own parsimony kernel (reported by Joan).
• Fix a minor issue when outputting mixture model name in model testing (reported by Sophie Abby).
• Fix a numerical problem with model testing (reported by Jana).
• For -t RANDOM initial model parameters will now be estimated on a parsimony tree (thanks to Jesse Breinholt for the report and suggestion).

Other changes:

• Windows AVX version (reported by Aaron Dickey) crashed due to a stack unalignment issue of TDM-GCC compiler (thanks to Agner Fog for suggestion). As a solution, all Windows binaries are now built with Clang.
• Updated the latest vectorclass library of Agner Fog.
• Print error instead of abort when applying branch tests for multifurcating trees.
• Implement EM algorithm for mixture+FreeRate model.
• Accept ~ as an unknown character in the alignment.
• Code optimization resulting in 5%-10% reduction in running time.

IQ-TREE version 1.3.10

October 16, 2015: We are pleased to announce version 1.3.10 with following changes:

• Support unlimited number of partitions for partition model analysis (no complaint about PLL_NUM_BRANCHES anymore).
• Improved support for mixture models.
• Fix a bug in ascertainment bias correction [+ASC] model causing incorrect branch length estimates (thanks to Marcus Teixeira for reporting it).
• Fix a rare bug when branch lengths are close to upper bound (thanks to Huaichun Wang for reporting it).

The parallel multicore version has download prefix iqtree-omp and the 32-bit version has prefix iqtree32.

IQ-TREE version 1.3.9

September 29, 2015: We are pleased to announce version 1.3.9 with following notable changes:

• Several bug fixes improving software stability. Special thanks to Xiaofan, Karen Meusemann, Jozsef Bakonyi, Renee, Peter Hosner for reporting bugs.
• A more stable expectation-maximization (EM) algorithm is implemented to optimize parameters of the FreeRate [+R] and LG4X model. Special thanks to Edward Susko, Thomas Wong and Lars Jermiin.
• Several new features (requested by users):
  • Option "-wbtl" to write bootstrap tree file (.ufboot) with branch lengths.
  • Option "-madd" to include mixture models into model selection procedure (e.g. "-madd LG4M,LG4X").
  • Option "-alrt 0" to perform the parametric approximate likelihood ratio (aLRT) branch test (Anisimova and Gascuel 2006).
  • Option "-abayes" to perform the parametric aBayes branch test (Anisimova et al. 2011).
• Maximum number of partitions is increased to 16384 instead of 1024.

v1.3.8: August 26, 2015: Bug-fix version 1.3.8 is released:

A bug introduced in 1.3.7 when using two options "-m TEST" (model testing) and -spp/-spj (edge-linked partition model) within one run.

IQ-TREE version 1.3.7

August 23, 2015: Version 1.3.7 is released with following changes:

• Fix a bug introduced in 1.3.6 for new model selection procedure (-m TESTNEW).
• Include L-BFGS-B algorithm (code taken from HAL_HAS package of Jayaswal et al. 2014; Syst Biol 63:726-742) as the default to estimate model parameters. L-BFGS-B performs better than the previous BFGS implementation, for example, when optimizing LG4X and FreeRate models.
• New option -suptag (used with -sup) when assigning support values from a set of input trees into a given tree, each tagged branch in the given tree will be assigned values of form support@tree1@tree2@...@treek, corresponding to the IDs of the input tree where this branch occurs (requested by Max Maronna).
• New option "-t RANDOM" to start tree search from a random starting tree (requested by Karen Meusemann).

IQ-TREE version 1.3.6

August 15, 2015: Version 1.3.6 is released with following substantial changes:

• An expectation-maximization (EM) algorithm is implemented to optimize weights of mixture models (thanks to Huaichun Wang, Andrew Roger, Edward Susko for reporting initial issue on local optimum and providing this solution). The EM algorithm guarantees convergence on global optimum and was described in Wang et al. (2008), BMC Evol. Biol. 8:331.
• Fix an issue in reading morphological alignments (thanks to Max Maronna for reporting this).
• +I+G is added back into the candidate model list of new model selection procedure (-m TESTNEW).
• The starting phase of tree search now includes a BIONJ into the candidate tree set, which is sometimes better than parsimony trees.
• Building from source code now generates dynamically linked binary. To build static binary, run cmake with e.g. "cmake -DIQTREE_FLAGS=static source_dir"