BAMM Changes¶

2.5.0¶

Nov 4, 2015

The prior distribution on the number of shifts is now specified with parameter expectedNumberOfShifts. This is simply the inverse of the previously-used parameter poissonRatePrior. BAMM control files will accept either parameter name but we find that expectedNumberOfShifts is more intuitive.
No more prior simulation in BAMM. The relevant BAMMtools functions now accept an argument for expectedNumberOfShifts and compute the full prior distribution from that information.
We made a major change to the algorithm for handling extinction probabilities $E(t)$ at internal nodes when the two branches descended from a node differ in their shift histories. We don’t consider the previous approach a bug per se, but have convinced ourselves that the new way has a better theoretical justification.

We have added a new function to BAMMtools, BAMMlikelihood, which is an R-based tool for replicating the calculations implemented in BAMM. This is described here.
Changes to BAMMtools rate plotting functions, including improved histograms, color break options, and more.
New functions, including generateControlFile (to generate BAMM control files from within R), plotPrior (to visualize the prior and posterior distributions on the number of rate shifts), and ratesHistogram (to visualize the distribution of rates across a phylogeny).
No more “branch specific Bayes factors” for distinguishing between core and non-core shifts. We have replaced this terminology with a related concept - the marginal odds ratio. Note that neither “branch specific Bayes factors” nor “marginal odds ratios” can be used for formal model selection; we have explained this in detail on our conceptual page about the interpretation of rate shifts.

the option validateEventConfiguration is an internal debugging option that should have had a default value of 0, but was instead set to 1 (this led to a bug in the Hastings ratio that would have affected a very small fraction of MCMC moves).

June 13, 2015

Caught a bug introduced during programming of fossil BAMM that allowed numerical overflow issues to occur to the extinction probability when E(t) approached 1.0 (this bug was not incorporated into the compiled version of BAMM that was distributed on the website but could have affected some analyses where users compiled BAMM code themselves).

May 26, 2015

Fixed bug in acceptance probability for MCMC moves that update the Poisson rate parameter. This is the most serious bug we have encountered in BAMM and would have magnified the effects of the prior on the posterior distribution of the number of shifts. If you previously used the default value of poissonRatePrior = 1.0, you will probably notice little effect on inference, but it may have had an impact in some cases. We are grateful to Cécile Ané and Bret Larget for helping us solve this.

September 5, 2014

Parallelism is now implemented using C++11 threads
- No dependency on OpenMP libraries
- Requires C++11 compiler (if building from source)
Provide a tool that helps choosing a deltaT value (available from GitHub), and it is documented in Metropolis coupled Markov chain Monte Carlo [(MC)3].
Add --version option in bamm and update “Usage” message
Fix bug in initialization of trait values

July 21, 2014

Add proposal to change the number of events by picking a random branch rather than a random location
- To use this proposal, add the setting updateRateEventNumberForBranch and give it a value greater than 0 (e.g., 0.1)
- It is recommended that the setting updateRateEventNumber be set to 0 when this new proposal is used
Fix log-likelihood calculation when loading event data or adding initial events

June 11, 2014

Metropolis-coupled MCMC
- Implementation from Altekar et al 2004 (Parallel Metropolis coupled Markov chain Monte Carlo for Bayesian phylogenetic inference)
- In the control file, specify the number of chains with numberOfChains, the temperature increment parameter (delta T) with deltaT, and the number of generations to wait for each chain swap proposal with swapPeriod
- Supports running chains in parallel with OpenMP
- Output acceptance information for chain swaps to the file specified by chainSwapFileName
Time-flip proposal
- Allows shift events to change from time-variable to time-constant and vice versa
- Enable this proposal with updateRateLambdaTimeMode for the speciation/extinction model type and with updateRateBetaTimeMode for the trait model type
Command-line options
- Change any setting at the command-line; for example, to change the number of generations to 1000, add --numberOfGenerations 1000 as a command-line option
- Any setting in the control file can be changed, and the new value specified at the command-line has precedence over that in the control file
Use different rate functions for the rate parameter k > 0 and for k < 0; when k > 0, use the symmetrical function of k < 0
Accept tree files with internal node names
Display better error messages related to reading a tree

March 5, 2014