Transfer bootstrap expectation vs. UFBoot

[davidcerny]

Hi everyone,

I'd like to calculate transfer bootstrap expectation (TBE) support values for my ML tree. This comment by @bqminh from the old IQ-TREE repo suggests that TBE shouldn't be used in conjunction with ultrafast bootstrap approximation (UFBoot), but the links provided in support of this statement point to a now-defunct Google group and do not resolve anymore. Unfortunately, it seems like the relevant information is not available anywhere else, or at least I have been unable to find it. Could someone please briefly summarize why it is inappropriate to combine these two ways of calculating branch support, and if there are any other relevant recommendations for using TBE within IQ-TREE?

[bqminh]

Hi there, the TBE (https://www.nature.com/articles/s41586-018-0043-0) was introduced in the context of standard Felsenstein's bootstrap and all the analysis and results published are associated with standard bootstrap. The reason why I'm reluctant to use TBE with UFBoot is because there is no study about the interpretation in this context, which is important because UFBoot values has a different meaning compared with standard bootstrap supports.

However, TBE can be already used with -b option (standard bootstrap) in IQ-TREE, simply by adding --tbe option.

[davidcerny]

Actually, I have one more question – hope that's okay. Let's say I want to annotate a previously inferred ML tree with TBE support values. I was hoping to do this as follows:

./iqtree2 -p partitions.nex -b 1000 -T AUTO --prefix SBS
./iqtree2 --tbe SBS.boottrees -sup ML.treefile --prefix annotated

The problem is that in the resulting tree file (annotated.suptree), the node annotations correspond to raw bootstrap frequencies rather than TBE values. Apparently, I can achieve what I want if I replace the second of the two commands above by the following call to Lemoine et al.'s (2018) original standalone program, booster:

./booster -i ML.treefile -b SBS.boottrees -o annotated.suptree

I'm just wondering if there's a way to do this from within IQ-TREE, too.

[bqminh]

Yes there is a trick, just repeat the first command adding -undo option (and --tbe):

./iqtree2 -p partitions.nex -b 1000 -T AUTO --prefix SBS --tbe -undo

Undo will pretend that the run hasn't finished, and it will load all checkpoints, until the point just before it finished previously, and then doing the TBE stuff. Personally I'm using undo when debugging the code....

[davidcerny]

Thank you, but I don't think this addresses the problem I had in mind – I probably explained it poorly. Your solution makes it possible to calculate the TBE values after the bootstrap run has finished, but it still only calculates them for the ML tree that was quietly inferred as a byproduct of that bootstrap run (SBS.treefile).

What I want to do is to calculate them for a separately inferred, user-provided ML tree (ML.treefile). Based on the manual and previous discussions, I was hoping to accomplish this with the -sup flag, but the -sup and --tbe flags don't seem to properly cooperate.

[bqminh]

I see, however -sup doesn't work with --tbe (as you suspected). So you would have to use booster.

[roblanf]

Not answering hte question (sorry!) but I think another good option here is one of the RAxML programs.

They implemented a faster and more memory efficient calculation of TBE in this paper: https://pubmed.ncbi.nlm.nih.gov/31755898/

Unless @bqminh puts me right, I think both IQ-TREE and booster use the older, slower version.

[bqminh]

yeah, correction, I think RAxML-NG included this library, not the original RAxML