LG4X model of protein substitution added to RAxML

[Alexandros Stamatakis]

Dear Users,

In addition to the LG4M model I have now also implemented the LG4X model
of protein substitution.

The models are described in this paper here:

http://mbe.oxfordjournals.org/content/29/10/2921.short

and are available in RAxML v7.4.5 on github.

Alexis

--
Alexandros (Alexis) Stamatakis

Research Group Leader, Heidelberg Institute for Theoretical Studies
Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology
Adjunct Professor, Dept. of Ecology and Evolutionary Biology, University
of Arizona at Tucson

www.exelixis-lab.org

[Ding]

Hi Alexis,

Bug report.

Mac OS 10.7.5

/raxmlHPC-PTHREADS-AVX -s data.phy -n data -f a -x 1234 -p 1234 -m PROTGAMMALG4XF -# autoMRE -T 8

#####

-181886.965809 -181838.490895

Assertion failed: (tr->likelihood >= initialLH), function optLG4X_Rate, file optimizeModel.c, line 2029.

Abort trap: 6

#####

/Ding

[Alexis]

Hi Ding I fixed this bug as well and pushed v7.4.6 to github.

Alexis

[Andrew Roger]

Hi  Alexis,

We were just running RAxML 8 on 4 x 2.8 Ghz AMD Opteron 6348 node (with 12 CPUs and 256GB RAM) and found that the LG4M model worked well, but the LG4X model seg. faulted with the
following error message:

raxmlHPC-PTHREADS-SSE3: optimizeModel.c:1410: optParamGeneric: Assertion `tr->likelihood >= initialLH' failed.
Aborted (core dumped)

Note that the same error occurred with the AVX version.  The command line was:
raxmlHPC-PTHREADS-SSE3 -T 24 -m PROTGAMMALG4X -n ASCT10-C.lg4xgamma -N 1 -s ASCT10-C.bmge.phy -p 1235

We also had problems running PROTGAMMAILG4X and PROTGRAMMAILG4M  (i.e., invariable sites option) with same error message.  I"m attaching the data file.

Note also in the manual you describe the 'automatic' model selection command as PROTAUTOGAMMA whereas it will only run with PROTGAMMAAUTO

Sincerely
Andrew Roger

[Alexandros Stamatakis]

Dear Andrew,

Thanks for reporting this, I'll try to have a look, but it might take
some time until I get to it.

Alexis

[Alexandros Stamatakis]

Dear Andrew,

> We were just running RAxML 8 on 4 x 2.8 Ghz AMD Opteron 6348 node (with 12
> CPUs and 256GB RAM) and found that the LG4M model worked well, but the LG4X
> model seg. faulted with the
> following error message:
>
> raxmlHPC-PTHREADS-SSE3: optimizeModel.c:1410: optParamGeneric: Assertion
> `tr->likelihood >= initialLH' failed.
> Aborted (core dumped)
>
> Note that the same error occurred with the AVX version. The command line
> was:
> raxmlHPC-PTHREADS-SSE3 -T 24 -m PROTGAMMALG4X -n ASCT10-C.lg4xgamma -N 1 -s
> ASCT10-C.bmge.phy -p 1235

This issue should be fixed now in version 8.0.13.

For the sake of correctness, this was not a segmentation fault, but an
assertion that failed.

You may be running RAxML with too many threads for the size of the
dataset by the way, please see previous discussions on this group.

> We also had problems running PROTGAMMAILG4X and PROTGRAMMAILG4M (i.e.,
> invariable sites option) with same error message. I"m attaching the data
> file.

It's not the same error message, the error message you get here is:

"Error: the LG4 substitution model does not work for proportion of
invariavble sites estimates!"

which means that +I is not implemented for the LG4M and LG4X models.

> Note also in the manual you describe the 'automatic' model selection
> command as PROTAUTOGAMMA whereas it will only run with PROTGAMMAAUTO

Thanks, I've updated the manual.

Alexis

>
> Sincerely
> Andrew Roger

[Andrew Roger]

Thanks Alexis.  Using 8.0.14 we have got the LG4X to work.  One thing that might be helpful is to discuss a bit more about the models in the manual and to specify that while the Q-matrices for both models are pre-computed, the rates/weights of the LG4X model are estimated by RAxML and that it is therefore a more complex model than LG4M.  I realize that you reference the Le et al. manuscript for the model, but maybe one or two sentences about what parts of these models are estimated versus what parts are pre-computed would be helpful, especially if the user wants to do model selection using AIC (or LRT) -- knowing how many additional parameters are being estimated relative to standard single-matrix models would be useful.

The only reason I say this is that reviewers are now usually insisting that some kind of model selection method was used to determine the model used in the analysis.  While the AUTO option is a great addition in this regard, it is likely that everyone analyzing amino acid data will start using LG4M and LG4X because they are usually WAY better fitting than single matrix models (actually any model that relaxes the assumption that the substitution dynamics are uniform across sites is WAY better fitting).  And I am now anticipating millions of posts to you asking you questions about numbers of parameters in LG4M and LG4X relative to the single matrix models...Just a thought.

[Andrew Roger]

In case its not clear - I did read the Le et al. paper and I do know that LG4X has 6 free parameters for its rates/weights versus the 1 alpha parameter for LG4M. 

[Alexandros Stamatakis]

I just updated the manual accordingly, if you score a tree with the -f e
option it will also automatically report the number of free parameters
for the AIC test. This has now also been corrected.

alexis

[nank...@ualberta.ca]

Hi Alexis,

When using the Cipres Science Gateway to run RAxML-HPC2 on XSEDE I keep receiving this error message.

raxmlHPC-HYBRID: optimizeModel.c:1605: optParamGeneric: Assertion `tr->likelihood >= initialLH' failed.
[gcn-4-71.sdsc.edu:mpi_rank_1][error_sighandler] Caught error: Aborted (signal 6)
[gcn-4-71.sdsc.edu:mpispawn_0][readline] Unexpected End-Of-File on file descriptor 5. MPI process died?
[gcn-4-71.sdsc.edu:mpispawn_0][mtpmi_processops] Error while reading PMI socket. MPI process died?
[gcn-4-71.sdsc.edu:mpispawn_0][child_handler] MPI process (rank: 1, pid: 4830) terminated with signal 6 -> abort job
[gcn-4-71.sdsc.edu:mpirun_rsh][process_mpispawn_connection] mpispawn_0 from node gcn-4-71 aborted: MPI process error (1)
[gcn-4-77.sdsc.edu:mpispawn_2][read_size] Unexpected End-Of-File on file descriptor 9. MPI process died?
[gcn-4-77.sdsc.edu:mpispawn_2][read_size] Unexpected End-Of-File on file descriptor 9. MPI process died?
[gcn-4-77.sdsc.edu:mpispawn_2][handle_mt_peer] Error while reading PMI socket. MPI process died?

The command line was 
/opt/raxml/bin/raxmlHPC-HYBRID -T 4 -n result -s infile.txt -m PROTGAMMALG4X -N 100 -p 12345 -b 12345

Would you be able to help with this?

Thanks,

Nerissa N

[Alexandros Stamatakis]

Hi Nerissa,

The RAxML version on CIPRES is typically lagging behind a couple of
version numbers w.r.t. the most current RAxML release.

Could you download the current RAxML version:

https://github.com/stamatak/standard-RAxML

and re-run your analysis with that?

If the error still occurs, please send me all input datasets and the
command line.

Cheers,

Alexis

> www.exelixis-lab.org <http://www.exelixis-lab.org>
>
> --
> You received this message because you are subscribed to the Google
> Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.

[Nicolas]

Hi, I am having the same problem with RAxML 8.2.11. The run gets through the bootstrap replicates and fails shortly after. I have tried with and without --no-bfgs, and also got the same error from both pthreads and mpi versions.

raxmlHPC-PTHREADS-SSE3 -s data.phy -f a -p 123 -x 123 -# 100 -T 20 -m PROTGAMMALG4X -n unpartitioned_ML

raxmlHPC-PTHREADS-SSE3: optimizeModel.c:1571: optParamGeneric: Assertion `tr->likelihood >= initialLH' failed.

Any help is much appreciated!

Best,

Nicolas

[Alexandros Stamatakis]

Dear Nicolas,

Could you please try if it works with the all new RAxML-NG:

https://github.com/amkozlov/raxml-ng

This is the up to date, better, faster, etc. version of RAxML.

Thank you,

Alexis

> www.exelixis-lab.org <http://www.exelixis-lab.org>
>
> --
> You received this message because you are subscribed to the Google
> Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.

--
Alexandros (Alexis) Stamatakis

Research Group Leader, Heidelberg Institute for Theoretical Studies
Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology

www.exelixis-lab.org