An estimation method that can use computer simulations to approximate maximum-likelihood estimates even when the likelihood function can not be evaluated directly. It can be applied whenever it is feasible to conduct many simulations, but works best when the data is approximately Poisson distributed. It was originally designed for demographic inference in evolutionary biology. It has optional support for conducting coalescent simulation using the 'coala' package.
Jaatha is an estimation method that uses computer simulations to produce maximum-likelihood estimates even when the likelihood function can not be evaluated directly. It can be applied whenever it is feasible to conduct many simulations, but works best when the data is at least approximately Poisson distributed.
Jaatha was originally designed for demographic inference in evolutionary biology. It has optional support for conducting coalescent simulation using the coala R package, but can also be used for different applications.
Jaatha can be installed from CRAN using the
Further help is provided using R's help system, in particular via
You can ask questions on jaatha's mailing list:
jaatha (at) googlegroups (dot) com
If you encounter problems when using jaatha, please
file a bug report or mail to
jaatha (at) googlegroups (dot) com.
Jaatha's original algorithm is described in the publication:
Jaatha: a fast composite-likelihood approach to estimate demographic parameters. Molecular Ecology 20(13):2709-23 (2011).
The revised version of the algorithm that is implemented in this package is described in:
L.A. Mathew, P.R. Staab, L.E. Rose and D. Metzler: Why to account for finite sites in population genetic studies and how to do this with Jaatha 2.0. Ecology and Evolution (2013).
Jaatha is developed openly on GitHub. Feel free to open an issue there if you encounter problems using Jaatha or have suggestions for future versions.
The current development version can be installed using:
boot_jaathawhich can be set to use more than one CPU core for each replica (#128).
folded_sums, which calculates the standard JSFS coarsening from an unpolarized spectrum (#130).
testthatis not available (#132).
zoom_in_stepsto the main jaatha
repetitionswas greater than one (#135). Thanks to Amaya Romero for reporting and locating this bug!.
verboseargument of the
1and repreat the runs with the new version if so (#124).
randominitialization method for using random starting positions (#104).
parsargument of an jaatha analysis and removed the corresponding arguments from
parallel::mclapplyis now set to
create_jaatha_model.coalmodelas functions (#109).
jaathaas a general method is now more prominent.
bootpackage for bootstrapping calculations.
Jaatha.confidenceIntevalsfunction. There often parameters different from the maximum likelihood estimates were used to calculate the confidence intervals (#15).
plyrfor converting JSFS into data frames because it is significantly faster.
epsilon. We now stop the refined search after the likelihood did not improve for 10 consecutive steps.
weightparameter to down-weight old simulations, because it had no measurable effect on the estimates.
rerunoption to initial & refined search