differences between RAxML Blackbox and XSEDE

[Martha Kandziora]

Hello,

I have a problem with my RAxML analyses on Cipres.
I ran one file once with the Blackbox and once with XSEDE (without applying additional/advanced settings under XSEDE).
For the two runs, I got two different best trees with a difference in the likelihood of about 400. Once, one of my species is placed in a supported clade. In the other analyses the whole clade is not supported and the species are placed basal.
Thus, I did some more runs to see if this obersavtion was reproducable. The species was always placed basal in the XSEDE analyses, while in the Blackbox it was placed in the supported clade. Furthermore, the different XSEDE runs I conducted resulted in different branch length for the different best trees, but all have a similar likelihood.
I tried to find some differences between the two and found out, that XSEDE can be run under GTRCAT and GTRGAMMA. I tried different settings, but still the same...

Can somebody explain this? I thought XSEDE is optimized for large datasets, but besides this the results should be similar or not?
Are there differences in the basic settings of the Blackbox and XSEDE?

Thanks for any comments or help,
Martha

[Mark Miller]

Hi Martha,

Thanks for reporting that issue. The two interfaces run on the same code, just one offers fewer options.

So in the end, they absolutely should give you the same answers. If you can provide me with the contents of the files

_jobinfo.txt for two runs using the same data but giving different answers, I will take a look.

Best,

Mark

[Mark Miller]

Dear Martha,

I had a look at your files, and did some experiments. Here is what I found:

If I run the RAXML_XSEDE interface with all options available, I found the trees were all just as you saw for RAxML BB,

except if the outgroup was included.

So:

raxmlHPC-HYBRID -s infile -M -n result -x 12345 -N 100 -q part  -c 25 -p 12345 -f a -m GTRCATI -o Syn_cappa  produced the weird tree where the species are basal.

But

raxmlHPC-HYBRID -s infile -M -n result -x 12345 -N 100 -q part  -c 25 -p 12345 -f a -m GTRCATI  produced the supported clade

raxmlHPC-HYBRID -s infile -n result -x 12345 -N 100 -q part  -c 25 -p 12345 -f a -m GTRCATI  produced the supported clade

and

raxmlHPC-HYBRID -s infile -M -n result -x 12345 -N automre -q part  -c 25 -p 12345 -f a -m GTRCATI produced the supported clade.

So if you neglect setting the outgroup, you will get your expected result, I believe.

I encourage you to report this on the RAxML Google group, and ask what it means.

Hope that helps.

Sincerely,