Topology moves under constraints

[zhangchicool]

MrBayes samples topologies under a set of constraints.

I’ve been running analyses with a set of 10 sequences sampled from major amniote groups (example dummy dataset attached). I'm interested in comparing inferences and marginal likelihoods calculated under a series of constraints. However, I recently noticed that depending on how the constraints are set up, MrBayes seems to have trouble sampling different topologies. More specifically, it seems to turn off all topology proposals in some instances, despite the fact that the topology is not fully specified (see the attached amnioteDummyConstraints.mb file). Weirdly, however, if I run a steppingstone analysis with these constraints, topology does change every once in a while, but not very often. I noticed this because I performed two replicate runs that started with different topologies and they sampled very different trees for most of their respective runs, ending with rather different estimates of the marginal likelihood.

This doesn’t seem to happen with all constraints, only certain combinations. For the attached data, if I only specify the “birdsPos” constraint then topology moves remain on.

All of these behaviors seem to occur in both versions 3.2.2 and 3.2.5.

example.zip (https://sourceforge.net/p/mrbayes/bugs/1627)

[igoryanovich]

Encountered the same problem.

Here is how to replicate it. Assume a .nex file has been loaded. When a topology with the trivial constraint "all" is imposed, showmoves demonstrates that the topology-changing moves have been disabled. When we apply a different constraint "some" after that, showmoves shows that the topology-changing moves are back.

I did not check whether this is specific to the combination of other priors - i.e. which ones in the example may be crucial for the error to appear.

constraint all = 1-.; constraint some = 2-.;

prset clockratepr=exponential(3e5); prset speciationpr=exp(1); prset extinctionpr=beta(1,1) nodeagepr=calibrated; prset brlenspr=clock:fossilization clockvarpr=TK02; prset speciationpr = uniform(0,1); prset extinctionpr = beta(1,1); prset tk02varpr = exponential(1); prset treeagepr = offsetexp(5000,20000); prset samplestrat=fossiltip; showmoves;

prset topologypr=constraints(all); showmoves;

prset topologypr=constraints(some); showmoves;