polytomies

[Stu Willis]

Hi Cecile et al.,

I was hoping to confirm a suspicion that I have had but have not seen explicitly addressed in the literature or here. It seems intuitive to me but I wanted to check before I put it in print.

If the gene tree inputs of a BCA have (soft) polytomies in them (e.g. 3+ OTUs share the same allele), MrBayes will create many random topologies for these loci, which mbsum will then summarize as having many low frequency topologies. The result should be lower concordance values in the BUCKy output, even if there is no hard conflict between the loci, correct?

So in this context, concordance could be interpreted as a good measure of positive support for relationships (or lack thereof), but a less accurate measure of true conflict, am I right?

Thanks for any thoughts!  Stu

[Cecile Ane]

Hi Stu,

The goal of BUCKy is to use information from other genes, which don't
have a soft polytomy, to inform the likely resolution for the gene that
does have a soft polytomy. So the answer to your question is no in
general: the result is not always low concordance values when there is
no hard conflict.

To illustrate with an example, I analyzed 30,000+ loci from Ebersberger
et al (2007) on 5 ape species (Ane 2010, book chapter). Two thirds of
the loci were either non-informative or non-clock like. Despite the
presence of so many polytomies for a large majority of loci, the results
were the similar when all loci of just the best 11,000+ loci were used.

However, with many taxa there is a problem in the 2-step approach of
BUCKy, which results in overestimated discordance: uncertainty (soft
polytomies) indeed results in lower concordance values with the current
version of BUCKy. I think there are other discussion threads that
discuss this issue with many taxa.

Cecile.

[Cymon Cox]

Dear Stu and Cecile,

Can I just come back to this (now rather old discussion)...

On Monday, 13 August 2012 02:23:47 UTC+1, Stu Willis wrote:

Hi Cecile et al.,

I was hoping to confirm a suspicion that I have had but have not seen explicitly addressed in the literature or here. It seems intuitive to me but I wanted to check before I put it in print.

If the gene tree inputs of a BCA have (soft) polytomies in them (e.g. 3+ OTUs share the same allele), MrBayes will create many random topologies for these loci, [SNIP]

Indeed MrBayes should randomly resolve that soft polytomy among the posterior samples, but it often doesn't due to the star-tree paradox (Lewis et al 2005).

So my question is, just how valid is BCA analysis for population-level data where many soft-polytomies can exist in any one locus tree?

(sorry if this has been addressed elsewhere, but I could find anything specific...)

Best regards, Cymon

[Cecile Ane]

Hi Cymon, 

Yes, MrBayes tends to give a high posterior probability for a single resolution of the polytomy and a small PP for all other resolutions, but we don’t expect this to happen on short alignments. We expect this behavior on long alignments,especially those obtained from concatenation. In the most common use of BUCKy, MrBayes is run on individual loci that are short enough that all resolutions are found with a positive posterior probability. 

If several loci have the same polytomy, and if the loci are long enough that MrBayes favors one resolution for each locus (star tree paradox), perhaps this is not so much of a problem because the favored resolution would vary from locus to locus. But that’s just guessing. I don’t know of any study that looked at this specific problem.

Cécile

[Cymon Cox]

Thanks for the reply Cecile, much appreciated,

Cymon

--
You received this message because you are subscribed to a topic in the Google Groups "BUCKy users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bucky-users/WEwI8SNjmiY/unsubscribe.
To unsubscribe from this group and all its topics, send an email to bucky-users...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

____________________________________________________________________

Cymon J. Cox