Introduction

This vignette briefly demonstrates multiple features of
babette, without going into much detail.
First, load the library:
This vignette shows how to:
- Let babetterun ‘BEAST2’
- Plot the posterior estimates
- Show the effective sample sizes (ESS)
- Show the summary statistics
- Plot the posterior phylogenies
In all cases, this is done for a short MCMC chain length of 10K:
inference_model <- create_test_inference_model()
Also, in all cases, we use the same BEAST2 options:
beast2_options <- create_beast2_options(verbose = TRUE)
 
Let babette run ‘BEAST2’
For an alignment, we’ll use a babette example
alignment.
if (is_beast2_installed()) {
  out <- bbt_run_from_model(
    fasta_filename = get_babette_path("anthus_aco_sub.fas"),
    inference_model = inference_model,
    beast2_options = beast2_options
  )
  bbt_delete_temp_files(
    inference_model = inference_model,
    beast2_options = beast2_options
  )
}
 
Plot the posterior estimates
if (is_beast2_installed()) {
  library(ggplot2)
  p <- ggplot(
    data = out$estimates,
    aes(x = Sample)
  )
  p + geom_line(aes(y = TreeHeight), color = "green")
  p + geom_line(aes(y = YuleModel), color = "red")
  p + geom_line(aes(y = birthRate), color = "blue")
}
 
Show the effective sample sizes (ESS)
Effective sample sizes, with 20% burn-in removed:
if (is_beast2_installed()) {
  traces <- remove_burn_ins(
    traces = out$estimates,
    burn_in_fraction = 0.2
  )
  esses <- t(
    calc_esses(
      traces,
      sample_interval = inference_model$mcmc$tracelog$log_every
    )
  )
  colnames(esses) <- "ESS"
  knitr::kable(esses)
}
For a reliable inference, use an ESS of at least 200.
 
Plot the posterior phylogenies
if (is_beast2_installed()) {
  plot_densitree(out$anthus_aco_sub_trees, width = 2)
}