Branch Lengths on majority rule consensus tree

[Fabricius Domingos]

Hi,

I have some doubts regarding the computation of a majority rule consensus tree (MR).

I have 2500 trees generated by the "same" genetic data (different phases), and I calculated BS using the RELL bootstrap option.

My doubts are:

1) Should I write BS values in each tree and then compute the MR, or should I compute MR using the 'best.trees' and then merge all BS files and draw to the MR tree? The second option seems more like the usual drawing BS values in a best scoring ML tree, but I'm really not sure.

2) Is there a way of computing a MR and also getting mean branch lengths and their standard deviation using RAxML? (like the sumt command would do in MrBayes)

Any help will be greatly appreciated!

Cheers,

  Fabricius

[Andre J. Aberer]

Hi Fabricius,

let me make sure, I understood you correctly: you have 2500 (or some
lower number) different datasets and computed best tree + bootstrap
support (using RELL)?

> 1) Should I write BS values in each tree and then compute the MR, or should
> I compute MR using the 'best.trees' and then merge all BS files and draw to
> the MR tree? The second option seems more like the usual drawing BS values
> in a best scoring ML tree, but I'm really not sure.

Well, both sound options sound legit, just like people are building
consensus trees and draw support onto a best-known tree (yielding 2
separate results from an analysis).

In case of 2) I would be careful though: simple concatenation of all BS
datasets makes datasets with more BS replicates have a stronger impact
on the BS support. If you follow down this path, you should choose an
identical number of trees (at random; use "sort -R | head -n number" )
for each dataset.

Be aware that for option 1), you will not obtain the weighted consensus
with RAxML (which you probably want), but the final consensus tree will
yield the number of consensus trees that support a specific branch. That
means if you compute the consensus of 2 consensus trees that both have a
bipartition with 70%, your final consensus tree will have 100% (because
the bipartition appears in both consensus trees). If you want the
weighted consensus tree, it makes sense to just concatenate all trees
(make sure that once more the number of trees per dataset is the same)
and compute the consensus of these.

> 2) Is there a way of computing a MR and also getting mean branch lengths
> and their standard deviation using RAxML? (like the sumt command would do
> in MrBayes)
>

As far as I know, it is not implemented in RAxML, since for BS
replicates people mostly inspect the topology. If you convert the tree
file into a Nexus format that is accepted by exabayes, you can use the
tools consense and extractBips for this task.

http://sco.h-its.org/exelixis/web/software/exabayes/index.html

--
Best regards,
Andre

[Fabricius Domingos]

Hi Andre,

Thank you very much for your answer.

> let me make sure, I understood you correctly: you have 2500 (or some 
> lower number) different datasets and computed best tree + bootstrap 
> support (using RELL)? 

Sorry that I was unclear about that in my first message: It's actually the same dataset, but with different phases of the heterozygous individuals (http://mbe.oxfordjournals.org/content/31/4/817). I created 2500 datasets (same data, same number of base pairs), and ran RAxML on each one of them, with exactly the same RAxML call. I'm assuming the same number of RELL BS replicates are created per run (1000 as far as I can see based on my results). So I think I should be fine concerning your comment in terms of the number os BS and number of trees. Based on your comments, I believe calculating the MRE and then drawing BS values (option 2) seems like the best option for me. Thanks for that!

I'll definitely have a try using ExaBayes then! Actually my post was motivated by the fact that MrBayes is crashing every time it reads my nexus tree (I'm getting an error message "Segmentation Fault: 11"). I hope I'll have more luck with exabayes. 

I might be getting ahead of myself here, but as far as I could understand from the exabayes manual, consense alone would do the job? (extractBips would extract ESS and PSRF values which would not make sense considering I'm using ML trees?)

Thanks again!

Best Regards,

   Fabricius

[Andre J. Aberer]

Hi Fabricius,

> Thank you very much for your answer.
>
>> let me make sure, I understood you correctly: you have 2500 (or some
>> lower number) different datasets and computed best tree + bootstrap
>> support (using RELL)?
>
> Sorry that I was unclear about that in my first message: It's actually the
> same dataset, but with different phases of the heterozygous individuals
> (http://mbe.oxfordjournals.org/content/31/4/817). I created 2500 datasets
> (same data, same number of base pairs), and ran RAxML on each one of them,
> with exactly the same RAxML call. I'm assuming the same number of RELL BS
> replicates are created per run (1000 as far as I can see based on my
> results). So I think I should be fine concerning your comment in terms of
> the number os BS and number of trees. Based on your comments, I believe
> calculating the MRE and then drawing BS values (option 2) seems like the
> best option for me. Thanks for that!
>

That sounds reasonable.

>
> I'll definitely have a try using ExaBayes then! Actually my post was
> motivated by the fact that MrBayes is crashing every time it reads my nexus
> tree (I'm getting an error message "Segmentation Fault: 11"). I hope I'll
> have more luck with exabayes.
>

Okay, although I fear, both programs have their very specific format
that needs to be obeyed. Maybe just try to replicate either of them.

>
> I might be getting ahead of myself here, but as far as I could
> understand from the exabayes manual, consense alone would do the job?
> (extractBips would extract ESS and PSRF values which would not make
> sense considering I'm using ML trees?)
>

Well, on the consensus tree, there is no standard deviation, I think,
but it listed in the output of extractBips.

--
Best regards,
Andre

--
Sent with my mu4e

[Fabricius Domingos]

Hi Andre,

> Okay, although I fear, both programs have their very specific format
> that needs to be obeyed. Maybe just try to replicate either of them.

Yes, I'll try and replicate exabayes nexus format.

> Well, on the consensus tree, there is no standard deviation, I think,
> but it listed in the output of extractBips.

Oh, I see! Thanks for pointing it out.
I'll let you know how it goes!

Thanks again!

Cheers,

Fabricius

> --
> You received this message because you are subscribed to a topic in the Google Groups "raxml" group.
> To unsubscribe from this topic, visit https://groups.google.com/d/topic/raxml/086eBbfS5sA/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to raxml+un...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Fabricius Domingos]

Hi Andre (and others),

I'm feeling quite stupid at the moment, but I tried everything that I could to load my RAxML trees in exabayes with no success.

I built the .nexus file based on a run using the example data of exabayes (attached: ExaBayes_topologies.myRun.0), and the only difference that I can notice between them is that exabayes trees use scientific notation and RAxML ones do not (all_trees_copy.nex - those are all "RAxML_bestTree.").

I'm running exabayes using the following: consense -f alltrees_copy.nex -n TEST

The error message that I'm getting is: expected >,<, got >?<

Would the scientific notation have this impact? If so, is there a way to remediate that (I have 2500 trees, can't change them all manually)? 

Or is there something else that I couldn't notice? Not sure where this 'comma' (>,<) that exabayes expects should be...

Thanks again for the help!

Best Regards,

   Fabricius

> To unsubscribe from this group and all its topics, send an email to raxml+unsubscribe@googlegroups.com.

[Andre J. Aberer]

Hi Fabricius,

the translation format in exabayes must be:

> translate
> 1 I3558MTR1,
> 2 I3559GRC2,

Furthermore, an unrooting was necessary (see attachment).

--
Best regards,
Andre

>> raxml+un...@googlegroups.com.

[Fabricius Domingos]

Thank you so much Andre!!!