Demographic Inference with Jaatha

Paul Staab

jaatha 3.2.5

When used for demographic inference, Jaatha supports using the package coala as simulation engine. If these packages do not suit your needs, it is of course also possible to use the normal interface described in the Introduction vignette.

Creating a Model

Jaatha automatically creates a simulation function, parameter ranges and summary statistics from a coala model. We can for example specify a simple isolation-with-migration model using par_ranges to mark parameters we want to estimate with Jaatha:

if(require("coala")) {  
   model <- coal_model(c(10, 15), 100) +
      feat_mutation(par_range("theta", 1, 10)) +
      feat_migration(par_range("m", 0, 3), symmetric = TRUE) +
      feat_pop_merge(par_range("t_split", 0.1, 2), 2, 1) + 
      feat_recombination(1) +
## Lade nötiges Paket: coala

We can now just pass this coala model to the create_jaatha_model function to convert it into a Jaatha model:

if(requireNamespace("coala")) {
     jaatha_model <- create_jaatha_model(model)
## A simulation takes less than a second

This uses coala for the simulations, gets the parameter ranges specified with par_range and uses summary statistics added to the model. Coala supports a wide range of models. Please refer to its documentation for more information.

Importing Data

You can use coala’s calc_sumstats_form_data function to calculate the summary statistic for genetic data. The output of this function can be directly passed on to create_jaatha_data.

Running Jaatha

From here on, you can estimate parameters using the jaatha as described in the introduction vignette.

If you are using a simulator that is writing temporary files to disk (e.g. ms, msms and seq-gen), please make sure that there is sufficient free space on your tempdir() to store the output of sim simulations per core that you use (arguments sim and cores in the jaatha function). Also, please make sure that your machine does not run out of memory. Both will lead to failtures during the estimation process. Reducing the number of cores reduces both the required memory and disk space at the cost of a longer runtime.