GeLL

• Likelihood - Using the method of Felsenstein 1981^[1]. Unobserved data can also be accounted for, per Felsenstein 1992^[2]. Confidence intervals for a parameter can also be calculated as discussed in Felsenstein 2003^[3].
• Ancestral Reconstruction
  • Marginal Reconstruction - Using the basic method of Yang, Kurma and Nei 1995^[4].
  • Joint Reconstruction - Using the methods of Pupko et al 2000^[5] and Pupko et al 2002^[6].
• Simulated Data - Using the standard methods as described, for example, in Yang 2006^[7].
• Non-reversible Processes - Using the method described by Boussau and Gouy 2006^[8].

GeLL comes in three different packages:

• Basic Package - This includes just the JAR file and the javadoc documentation
• Source Package - This additionally includes the source code
• Test Package - This additionally includes both the source code and test classes

GeLL comes with full javadoc documentation in the javadoc directory here.

GeLL includes a driver that should be capable of doing most basic computations. This is documented in the driver documentation here.

The GeLL packages also include an example program. As well as the source code file an annotated version of the source is here. If you are thinking of writing your own program using the GeLL library it is recommended you take a look at this example.

The core capability of GeLL is the ability to define and use new models. As such how models are defined in covered in more details here. Also included is a more in-depth discussion of how parameters are defined and constrained.

A description of the test cases is included here in the test package.

• Gaps
  Gaps can be dealt with in two ways. They can be treated as an extra state in which case the extra state should simply be added to the model. To sum across all states, like many models do, simply define the gap character as being an ambiguous character that could be any one of the states.

• Eigendecomposition
  Eigendecompositions may not always converge and many not throw an error. The eigendecomposition code is taken from JAMA and has not been debugged. This seems to especially be the case for gene family size models.
• Parameters (1)
  There is an issue where using the same Parameters object in two or more calls to different Calculator objects will cause an error. This can be solved by passing a cloned copy of the parameters in the second
• Parameters (2)
  The way branch lengths are treated as parameters is a little odd and could well cause confusion. It is intended to overhaul this in a future version.
• Exceptions
  Some exceptions are thrown in circumstances where it is known that the exception should not occur, normally because we know that the calling class has already checked (either implicitly or explicitly) that the exception can't occur. It is intended to fix this in a later release.
• Threaded Computation - Likelihood
  Currently threaded computation is always used for likelihood computation. It is known that this can be slower for small trees and small state-space models. However, it has not been yet been determined under what conditions threaded calculation is slower so, for the moment, it is always used.
• Threaded Computation - Ancestral reconstruction
  At the moment ancestral reconstruction is not threaded.
• Optimising No Parameters
  Trying to optimise a parameters object with no optimisable parameters (i.e. they are all fixed values) can result in an infinite loop.

• First release

• Not publicly release due to memory problems
• Added support for model hypothesis testing
• Allow sites to be in different class

• Not (quite) backwards compatible although only minimal changes should be necessary
• Changes to internal storage structures to sort memory problems
• Large amounts of code optimisation

• Not (quite) backwards compatible although only minimal changes should be necessary
• Added Conjugate Gradient search
• Added an ID field to Site
• Added split support to Tree and also the ability to calculate various distances between trees (RF, weighted RF and branch score).
• Minor Changes:
  • Optimiser don't store node likelihood until after optimisation is complete to save memory.
  • Attempt to make calculating distributions using the "repeated" method more stable.
  • Fixed bugged in test for quasi-stationary distribution convergence.
  • Parameters toString method now only outputs estimated parameters by default.
  • Ability to write sequence alignments in Phylip Alignments in PAML format, i.e. taxa names have a maximum of 15 characters. This is buggy as names are simply truncated and duplicated names are not checked for.
  • Fix to PossibleSettings which meant settings in no group did not work properly.
  • Parameter now throws an error if an attempt is made to set it to an invalid value.
  • Additional error checking in likelihood calculations to check for probabilities greater than one and other errors.
  • Various updates to throw statements caused by the above.
  • Moved the storing of whether matrices need to be recalculated from Parameters to RateCategory. This is more inituative as the what parameters are in the rate category decides when changes to parameters needs matrix recalculations. Should, hopefully, be largely transparent.

• Not (quite) backwards compatible although only minimal changes should be necessary
• Added the ability to work with very small likelihoods. This is optional as it results in a significant slow down. There is a slight performance decrease even when the small option is not being used.
• Added the ability to use the FitzJohn (2009) method as the root distribution.
• Major changes to how calculators etc are structured to allow more unusual ways of calculating likelihoods.
• Removed code to do with Constraints as it was implemented in a very odd way and was no longer needed.
• Added BDIE (Birth-Death-Innovation-Extinction) and Birth-Death (with no zero state) models to the duplication model factory.
• Minor Changes:
  • Removed ArrayMap as it was not providing any speed increase so was just complicating the code.
  • Removed ToDoubleHashMap, again as it was complicating the code.
  • Removed deprecated SequenceAlignment. PhylipAlignment should be used instead.
  • Removed unused MapUtils.
  • Added an additional example executable that implements the model of chromosome evolution of Mayrose et al (2009). This example shows how to implement a new model.
  • DNAModel factory now have two options to create models. One functions takes in and updates the parameters list with relevant parameters whereas the other just creates the model.
  • Code now automatically checks whether a parameter is in the matrix so it no longer part of the Parameter. Relevant constructors removed.
  • Added an ability to generate a random unrooted tree.

GeLL is avaliable under GPL v3.

The EigenvalueDecomposition and LUDecomposition classes are from JAMA and as such are in the public domain.

The test package contains both executables and example datasets from PAML. This example dataset is also used for the example execution using Example.java. The applicable license for these components is as follows:

The latest stable release of the library is available at http://phylo.bio.ku.edu/GeLL, while the latest developmental version is available at http://github.com/danielmoney/GeLL.

The author is contactable at daniel.money@dal.ca

1. Felsenstein, J. 1981. Evolutionary trees from DNA sequences: A maximum likelihood approach. Journal of Molecular Evolution, 17:368–376.
2. Felsenstein, J. 1992. Phylogenies from restriction sites: A Maximum-Likelihood approach. Evolution, 46:159–173.
3. Felsenstein, J. 2003. Inferring phylogenies. Sinauer Associates Sunderland, Mass., USA.
4. Yang, Z., S. Kumar, and M. Nei. 1995. A new method of inference of ancestral nucleotide and amino acid sequences. Genetics, 141:1641–1650.
5. Pupko, T., I. Pe, R. Shamir, and D. Graur. 2000. A fast algorithm for joint reconstruction of ancestral amino acid sequences. Molecular Biology and Evolution, 17:890–896.
6. Pupko, T., I. Pe’er, M. Hasegawa, D. Graur, and N. Friedman. 2002. A branchand- bound algorithm for the inference of ancestral amino-acid sequences when the replacement rate varies among sites: Application to the evolution of five gene families. Bioinformatics, 18:1116–1123.
7. Yang, Z. 2006. Computational molecular evolution. Oxford University Press, USA.
8. Boussau, B. and Gouy, M. 2006 Efficient Likelihood Computations with Nonreversible Models of Evolution. Systematic Biology, 55:756-768.