MrBayes

Is there a difference between using MrBayes with mpirun or mpiexec? On 
our cluster we use MVAPICH and Mpiexec for mvapich, mpirun is disabled. 
But every time i start mb through PBS using 1,2,...8 cores , looks like 
mb sees only one available processors , and despite the spawning of 
1,2...8 processes, they all share 1 CPU core only.

Hi MrBayes users,

I am running MrBayes on long aligmnents (nchar = 16955) of few (12) taxa,
and I find the program behaving differently than with my usual datasets of
more (3x to 6x) taxa and shorter (1/10 to 1/30) alignments.  In particular,
with my long alignments, I find the average standard deviation of split
frequencies converges VERY quickly to below 0.01, and on these short runs
(maybe inevitable with very short runs?), the plot generated by sump of the
generation versus the log probability of the data shows values not evenly
distributed (points have an obvious trend).  (BTW- These runs are amino
acid, using "mixed" model, always resulting in heavy use of Wag matrix.)
Do long alignments result in really low values for average standard
deviation of split frequencies, just because of their length? (I frequently
get convergence to below 0.000000 in 10,000 generations.)
If I continue runs long after convergence to below 0.01, or lower, is the
data generated more or less meaningful than data generated on a run which is
more slowly converging towards 0.01?
Thanks for any advice.
Beatrice

I think it is because you have a reduced taxon sampling (12 taxa). I
work with a similar alignments but >50 taxa, and your kind of
convergence is just a mirage!

best
Omar

2010/6/15 Beatrice Parker <beatrice.d.parker@...>:
> Hi MrBayes users,
>
> I am running MrBayes on long aligmnents (nchar = 16955) of few (12) taxa,
> and I find the program behaving differently than with my usual datasets of
> more (3x to 6x) taxa and shorter (1/10 to 1/30) alignments.  In particular,
> with my long alignments, I find the average standard deviation of split
> frequencies converges VERY quickly to below 0.01, and on these short runs
> (maybe inevitable with very short runs?), the plot generated by sump of the
> generation versus the log probability of the data shows values not evenly
> distributed (points have an obvious trend).  (BTW- These runs are amino
> acid, using "mixed" model, always resulting in heavy use of Wag matrix.)
>
> Do long alignments result in really low values for average standard
> deviation of split frequencies, just because of their length? (I frequently
> get convergence to below 0.000000 in 10,000 generations.)
>
> If I continue runs long after convergence to below 0.01, or lower, is the
> data generated more or less meaningful than data generated on a run which is
> more slowly converging towards 0.01?
>
> Thanks for any advice.
> Beatrice
> ------------------------------------------------------------------------------
> ThinkGeek and WIRED's GeekDad team up for the Ultimate
> GeekDad Father's Day Giveaway. ONE MASSIVE PRIZE to the
> lucky parental unit.  See the prize list and enter to win:
> http://p.sf.net/sfu/thinkgeek-promo
> _______________________________________________
> Mrbayes-users mailing list
> Mrbayes-users@...
> https://lists.sourceforge.net/lists/listinfo/mrbayes-users
>



-- 
Omar Rota Stabelli
Pisani Evolutionary Biology Lab, Department of Biology,
National University of Ireland, Maynooth, Co. Kildare, Ireland
       --------------------------------
What if the very idea of growth accumulating riches, destroying the
environment and worsening social inequality is a trap? Maybe we need
to aim to create a society that is based on quality not quantity, on
cooperation and not competition.  Serge Latouche
       --------------------------------

Deviation of split frequencies simply measures the similarity of topologies across the two runs.  With twelve taxa and lots of data you are likely to converge on identical/nearly identical trees rapidly and to remain at that optimum.  This tells you little or nothing about convergence in other parameters.


On 6/16/10 4:52 AM, "Omar Rota Stabelli" <omar42@...> wrote:

I think it is because you have a reduced taxon sampling (12 taxa). I
work with a similar alignments but >50 taxa, and your kind of
convergence is just a mirage!

best
Omar

2010/6/15 Beatrice Parker <beatrice.d.parker@...>:
> Hi MrBayes users,
>
> I am running MrBayes on long aligmnents (nchar = 16955) of few (12) taxa,
> and I find the program behaving differently than with my usual datasets of
> more (3x to 6x) taxa and shorter (1/10 to 1/30) alignments.  In particular,
> with my long alignments, I find the average standard deviation of split
> frequencies converges VERY quickly to below 0.01, and on these short runs
> (maybe inevitable with very short runs?), the plot generated by sump of the
> generation versus the log probability of the data shows values not evenly
> distributed (points have an obvious trend).  (BTW- These runs are amino
> acid, using "mixed" model, always resulting in heavy use of Wag matrix.)
>
> Do long alignments result in really low values for average standard
> deviation of split frequencies, just because of their length? (I frequently
> get convergence to below 0.000000 in 10,000 generations.)
>
> If I continue runs long after convergence to below 0.01, or lower, is the
> data generated more or less meaningful than data generated on a run which is
> more slowly converging towards 0.01?
>
> Thanks for any advice.
> Beatrice
> ------------------------------------------------------------------------------
> ThinkGeek and WIRED's GeekDad team up for the Ultimate
> GeekDad Father's Day Giveaway. ONE MASSIVE PRIZE to the
> lucky parental unit.  See the prize list and enter to win:
> http://p.sf.net/sfu/thinkgeek-promo
> _______________________________________________
> Mrbayes-users mailing list
> Mrbayes-users@...
> https://lists.sourceforge.net/lists/listinfo/mrbayes-users
>



--
Omar Rota Stabelli
Pisani Evolutionary Biology Lab, Department of Biology,
National University of Ireland, Maynooth, Co. Kildare, Ireland
       --------------------------------
What if the very idea of growth accumulating riches, destroying the
environment and worsening social inequality is a trap? Maybe we need
to aim to create a society that is based on quality not quantity, on
cooperation and not competition.  Serge Latouche
       --------------------------------

------------------------------------------------------------------------------
ThinkGeek and WIRED's GeekDad team up for the Ultimate
GeekDad Father's Day Giveaway. ONE MASSIVE PRIZE to the
lucky parental unit.  See the prize list and enter to win:
http://p.sf.net/sfu/thinkgeek-promo
_______________________________________________
Mrbayes-users mailing list
Mrbayes-users@...
https://lists.sourceforge.net/lists/listinfo/mrbayes-users


______________________________________________________________
Ted R. Schultz, Ph.D.
Research Entomologist
Department of Entomology
Smithsonian Institution
POB 37012
NHB, CE516, MRC 188
Washington, DC 20013-7012
U.S.A.

Use ABOVE ADDRESS for U.S. Postal Service ONLY.
Mail sent via U.S. Postal Service to our street address (below) is detoured to Ohio to get fried by gamma radiation (similar to being baked in a 350°F oven for 12 hours).

E-mail: schultzt@...
Web pages:
http://entomology.si.edu/StaffPages/schultz.htm
http://entomology.si.edu/SIAntLab.html
http://www.mnh.si.edu/exhibits/ants/Moffett/ants/index.html
Phone (Entomology main office): 00-1-202-633-1033
Fax: 00-1-202-786-2894
Phone (Schultz office): 00-1-202-633-0999
Phone (Ant Laboratory): 00-1-202-633-1001
Phone (Molecular Entomology Lab): 00-01-301-238-3160

Use BELOW ADDRESS for ALL PRIVATE CARRIERS (e.g, Fed Ex, UPS, DHL).
DO NOT use address below for U.S. Postal Service unless you
want your package to get FRIED by gamma radiation, which,
incidentally, is really good for converting DNA molecules into
single nucleotides.

Ted Schultz
Natural History Bldg., CE516
10th & Constitution Aves. NW
Washington, DC 20560-0188
U.S.A.

Phone (NMNH Shipping Office): 00-1-202-633-1637

Thank you so much for the advice.

I *have* 46 taxa, I'll use them all.  Do you think that's enough to
significantly improve the validity of my results?

POSSIBLY HARDER / WEIRDER QUESTION:
The alignments are concatenated protein sequences-- should I separate them
into individual protein alignments, run the individual mcmc simulations, and
then manually combine all the .p and .t files (manually removing a burnin
from each one, I guess) and do one sump and sumt on the aggregated total
results (setting burnin at that point to 0, I guess)?  If that works,
would the topology and bipartition probability values be more meaningful?
(Is manually removing the burnin from the .t and .p files equivalent to
setting burnin during the sump and sumt commands?)

Thanks in advance
Beatrice

On Wed, Jun 16, 2010 at 8:14 AM, Schultz, Ted <SCHULTZT@...> wrote:

> Deviation of split frequencies simply measures the similarity of topologies
> across the two runs.  With twelve taxa and lots of data you are likely to
> converge on identical/nearly identical trees rapidly and to remain at that
> optimum.  This tells you little or nothing about convergence in other
> parameters.
>
>
> On 6/16/10 4:52 AM, "Omar Rota Stabelli" <omar42@...> wrote:
>
> I think it is because you have a reduced taxon sampling (12 taxa). I
> work with a similar alignments but >50 taxa, and your kind of
> convergence is just a mirage!
>
> best
> Omar
>
> 2010/6/15 Beatrice Parker <beatrice.d.parker@...>:
> > Hi MrBayes users,
> >
> > I am running MrBayes on long aligmnents (nchar = 16955) of few (12) taxa,
> > and I find the program behaving differently than with my usual datasets
> of
> > more (3x to 6x) taxa and shorter (1/10 to 1/30) alignments.  In
> particular,
> > with my long alignments, I find the average standard deviation of split
> > frequencies converges VERY quickly to below 0.01, and on these short runs
> > (maybe inevitable with very short runs?), the plot generated by sump of
> the
> > generation versus the log probability of the data shows values not evenly
> > distributed (points have an obvious trend).  (BTW- These runs are amino
> > acid, using "mixed" model, always resulting in heavy use of Wag matrix.)
> >
> > Do long alignments result in really low values for average standard
> > deviation of split frequencies, just because of their length? (I
> frequently
> > get convergence to below 0.000000 in 10,000 generations.)
> >
> > If I continue runs long after convergence to below 0.01, or lower, is the
> > data generated more or less meaningful than data generated on a run which
> is
> > more slowly converging towards 0.01?
> >
> > Thanks for any advice.
> > Beatrice
>

On Wed, 16 Jun 2010, Beatrice Parker wrote:

> The alignments are concatenated protein sequences-- should I separate them
> into individual protein alignments, run the individual mcmc simulations,

The most obvious thing to do is to use a partitioned model on the 
concatenated alignment, if it is not already the case.

It is of course a good idea to run separate analyses also, and then use 
some consensus or supertree method on the resulting trees.


> and then manually combine all the .p and .t files (manually removing a burnin
> from each one, I guess) and do one sump and sumt on the aggregated total
> results (setting burnin at that point to 0, I guess)?

If you want to combine the .t and .p files from runs on separate data, you 
will have to discard as burnin what is needed for the slowest of the runs.
If you ensure that all the separate runs have the same length, then it is 
probably not necessary to manually remove the burnin period, just do it 
the normal way, only using .t and .p files coming from separate analyses.


> If that works, would the topology and bipartition probability values be 
> more meaningful?

Producing trees directly from the combination of the .t files might amount 
to some sort of majority-rule consensus. I'm not sure whether the 
"posterior probabilities" obtained using such a "hack" would still have 
their original meaning, but I suppose they would still give an idea of the 
probabilities of the bipartitions.


> (Is manually removing the burnin from the .t and .p files equivalent to
> setting burnin during the sump and sumt commands?)

I suppose so.


Blaise Li

Hi

Loading a mixed data file with 2 DNA partitions and 55 protein 
partitions ~ 14000 characters total gives same 'stack smashing' error 
with 64-bit compile using Vladimir's patches . . .

       Compressing data matrix for division 57
       Division 57 has 897 unique site patterns
*** stack smashing detected ***: ./mb_64 terminated
======= Backtrace: =========
/lib/libc.so.6(__fortify_fail+0x37)[0x7f1590d751d7]
/lib/libc.so.6(__fortify_fail+0x0)[0x7f1590d751a0]
./mb_64[0x4608e5]
./mb_64[0x4951e5]
./mb_64[0x41f804]
./mb_64[0x42011d]
./mb_64[0x41f804]
./mb_64[0x405052]
./mb_64[0x405557]
/lib/libc.so.6(__libc_start_main+0xfd)[0x7f1590c94c4d]
./mb_64[0x4019c9]


Too many unlinked partitions maybe . .

On 06/09/2010 08:26 PM, Mariana Mondragon wrote:
> I just tried running an analysis with MrBayes and experienced the same
> problem that has been reported as a bug on 06-06-2010. I get exactly the
> same error that is indicated in this report. Although I gave the correct
> sumt commands included in the file with the alignment, these were not
> executed. When I tried them on the command line of the program i got the
> following output, please see below. I am running the program in a cluster
> running Linux SLES 10.

Hi Mariana,

I guess the problem could be that you are running MrBayes in 64-bit mode
but without the patches to make it 64-bit safe (and without activating
the patch).

See my instructions at

http://technical.bestgrid.org/index.php/Bioinformatics_applications_at_University_of_Canterbury_HPC#Installing_MrBayes

and recompile MrBayes accordingly.


Cheers,
Vladimir

-- 
Vladimir Mencl, Ph.D.
E-Research Services and Systems Consultant
BlueFern Supercomputing Services
University of Canterbury
Private Bag 4800
Christchurch 8140
New Zealand

http://www.bluefern.canterbury.ac.nz
mailto:vladimir.mencl@...
Phone: +64 3 364 3012
Mobile: +64 21 997 352
Fax: +64 3 364 2332

Many thanks Vladimir,

We patched the MrBayes and now it is running. In a few hours it will get to
the critical point where it had some trouble.

Bests

Mariana

On Wed, Jun 9, 2010 at 10:34 PM, Vladimir Mencl <
vladimir.mencl@...> wrote:

> On 06/09/2010 08:26 PM, Mariana Mondragon wrote:
> > I just tried running an analysis with MrBayes and experienced the same
> > problem that has been reported as a bug on 06-06-2010. I get exactly the
> > same error that is indicated in this report. Although I gave the correct
> > sumt commands included in the file with the alignment, these were not
> > executed. When I tried them on the command line of the program i got the
> > following output, please see below. I am running the program in a cluster
> > running Linux SLES 10.
>
> Hi Mariana,
>
> I guess the problem could be that you are running MrBayes in 64-bit mode
> but without the patches to make it 64-bit safe (and without activating
> the patch).
>
> See my instructions at
>
>
> http://technical.bestgrid.org/index.php/Bioinformatics_applications_at_University_of_Canterbury_HPC#Installing_MrBayes
>
> and recompile MrBayes accordingly.
>
>
> Cheers,
> Vladimir
>
> --
> Vladimir Mencl, Ph.D.
> E-Research Services and Systems Consultant
> BlueFern Supercomputing Services
> University of Canterbury
> Private Bag 4800
> Christchurch 8140
> New Zealand
>
> http://www.bluefern.canterbury.ac.nz
> mailto:vladimir.mencl@...
> Phone: +64 3 364 3012
> Mobile: +64 21 997 352
> Fax: +64 3 364 2332
>
>


-- 
Mariana Mondragon-Palomino, Ph.D.

Cell Biology & Plant Biochemistry
University of Regensburg
Universitaetsstrasse 31
93053  Regensburg
Germany

Telephone (office) +49 941-943 3009
               (lab)    +49 941-943 3042
Fax                     +49 941-943 3352

Hi!

Which version of MrB is this? Did you use the very same dataset file to
generate the trees in the first place? Have you checked so that taxon
names are not repeated in the data matrix or the tree file? Do you use
any weird non-alphanumerical symbols in the taxon names?

/Andreas

On 2010-06-06 20:34, Antoine wrote:
> Hi
> I ran an analysis but when comes the sumt step
> i got the following msg: Taxon name 9 already used in tree
> error in command sumt"
> I already tried a cuple of things to go around the problem but... failed...
> Is there anyone who already had this problem before?
> Antoine
>
> ------------------------------------------------------------------------------
> ThinkGeek and WIRED's GeekDad team up for the Ultimate 
> GeekDad Father's Day Giveaway. ONE MASSIVE PRIZE to the 
> lucky parental unit.  See the prize list and enter to win: 
> http://p.sf.net/sfu/thinkgeek-promo
> _______________________________________________
> Mrbayes-users mailing list
> Mrbayes-users@...
> https://lists.sourceforge.net/lists/listinfo/mrbayes-users
>

Hi Andreas,
> Hi!
>
> Which version of MrB is this? 
3.1.2
> Did you use the very same dataset file to
> generate the trees in the first place?
no. I got the .t and . p files from the cluster. Then I ran a quick job 
on my laptop and when it reached the sump sumt step I replaced the small 
.t .p files by the good ones in order to produce the . con .tparts ect
I did that several times and never had such problem before....

>  Have you checked so that taxon
> names are not repeated in the data matrix or the tree file? Do you use
> any weird non-alphanumerical symbols in the taxon names?
>   
well I don't think so, I just checked it out and one taxon as a "." but 
it is the #59     not the #9 and sump works fine and anyway the error 
msg is "

Taxon name 9 already used in tree
error in command sumt"

Any clue?
Cheers,
Antoine

> /Andreas
>
> On 2010-06-06 20:34, Antoine wrote:
>   
>> Hi
>> I ran an analysis but when comes the sumt step
>> i got the following msg: Taxon name 9 already used in tree
>> error in command sumt"
>> I already tried a cuple of things to go around the problem but... failed...
>> Is there anyone who already had this problem before?
>> Antoine
>>
>> ------------------------------------------------------------------------------
>> ThinkGeek and WIRED's GeekDad team up for the Ultimate 
>> GeekDad Father's Day Giveaway. ONE MASSIVE PRIZE to the 
>> lucky parental unit.  See the prize list and enter to win: 
>> http://p.sf.net/sfu/thinkgeek-promo
>> _______________________________________________
>> Mrbayes-users mailing list
>> Mrbayes-users@...
>> https://lists.sourceforge.net/lists/listinfo/mrbayes-users
>>   
>>     
>
>
>