pmsf

[Richard Benji]

Hello

I have a question about the use of PMSF. In the paper of Wang 2018 and on the GitHub page (https://github.com/Cibiv/IQ-TREE/wiki/Complex-Models), you demonstrate that PMSF is (to do simple) as efficient as Cxx or CAT+GTR, but faster.

You show that with LG+PMSF+Γ allows us to be faster for than LG+C60+F+G and LG+C20+F+G.




My question is :

If I do LG+C60+F+G with a guide tree, will it be faster than a LG+C60+F+G ? Is it a way to use PMSF ?




I think there is something I don't understand well in this Table from your GitHub and


Wang paper :

In this Table


Models	CPU time	Wall-clock time	RAM usage
LG+F+G	43h:38m:23s	3h:37m:23s	1.8 GB
LG+C20+F+G	584h:25m:29s	46h:39m:06s	38.8 GB
LG+C60+F+G	1502h:25m:31s	125h:15m:29s	112.8 GB
LG+PMSF+G	73h:30m:37s	5h:7m:27s	2.2 GB




What PMSF refer to ? Do I need to replace PMSF by a model (for ex. C60 or C20 ?) ? Are they use a Cxx instead with a guide tree ? Because I don't understand what model they use to save all this time in the paper.





I have 500 fasta of 121 species and 35 000 positions. I tried with the following line command :





for f in *-modified.fasta; do echo "iqtree-mpi -s $f -m LG+C60+F+G -ft replica-1-concat.fasta-modified.treefile -nt 8 -mem 60G"; done > all-jobs.cmds




I succeeded in getting 15 trees (2 days of calculation, and it seems much compared to your data set of Brown who have results in 5h with 68 sequences, 43615 amino-acid sites, 16 CPU... and I use 32 CPU).




Plus, for the 75 others, I had this erreur (not for the 15 first fasta, I don't understand) :





ERROR: STACK TRACE FOR DEBUGGING:
ERROR: 1   double* aligned_alloc<double>(unsigned long)
ERROR: 2   PhyloTree::initializeAllPartialLh(int&, int&, PhyloNode*, PhyloNode*)
ERROR: 3   PhyloTree::initializeAllPartialLh()
ERROR: 4   computeSiteFrequencyModel(Params&, Alignment*)
ERROR: 5   runPhyloAnalysis(Params&, Checkpoint*)
ERROR: 6   main()
ERROR: 7   __libc_start_main()
ERROR: Not enough memory, allocation of 59743927072 bytes failed (bad_alloc)



Do you think my parameters -nt 8 -mem 60G are good ?





According to you, using PMFS, what would be the most pertinent ? What we want is the equivalent of a C60, without having as much computing time.




Best regards




Richard GOUY


[Richard Benji]

I forgot to mentioned that my guide trees results from standard LG+C60.

[Minh Bui]

Hi Richard,

On 13 Apr 2020, at 2:37 pm, Richard Benji <rgo...@gmail.com> wrote:

Hello

I have a question about the use of PMSF. In the paper of Wang 2018 and on the GitHub page (https://github.com/Cibiv/IQ-TREE/wiki/Complex-Models), you demonstrate that PMSF is (to do simple) as efficient as Cxx or CAT+GTR, but faster.

You show that with LG+PMSF+Γ allows us to be faster for than LG+C60+F+G and LG+C20+F+G.




My question is :

If I do LG+C60+F+G with a guide tree, will it be faster than a LG+C60+F+G ? Is it a way to use PMSF ?

No, you should use LG+F+G to make the guide tree. Otherwise there is no runtime benefit, of course.




I think there is something I don't understand well in this Table from your GitHub and


Wang paper :

In this Table


Models	CPU time	Wall-clock time	RAM usage
LG+F+G	43h:38m:23s	3h:37m:23s	1.8 GB
LG+C20+F+G	584h:25m:29s	46h:39m:06s	38.8 GB
LG+C60+F+G	1502h:25m:31s	125h:15m:29s	112.8 GB
LG+PMSF+G	73h:30m:37s	5h:7m:27s	2.2 GB




What PMSF refer to ?

PMSF is abbreviation for Posterior mean site frequency. But please note that in the command line, you should use -m LG+C60+F+G -ft <GUIDE_TREE_FILE>. So a typical analysis has 2 steps:

1. Build a tree guide:

iqtree -s ALN_FILE -m LG+F+G -pre guidetree

2. Use the .treefile from the above run to do this approximation:

iqtree -s ALN_FILE -m LG+F+C60+G -ft guidetree.treefile ...

Do I need to replace PMSF by a model (for ex. C60 or C20 ?) ? Are they use a Cxx instead with a guide tree ? Because I don't understand what model they use to save all this time in the paper.


See above.




I have 500 fasta of 121 species and 35 000 positions. I tried with the following line command :





for f in *-modified.fasta; do echo "iqtree-mpi -s $f -m LG+C60+F+G -ft replica-1-concat.fasta-modified.treefile -nt 8 -mem 60G"; done > all-jobs.cmds




I succeeded in getting 15 trees (2 days of calculation, and it seems much compared to your data set of Brown who have results in 5h with 68 sequences, 43615 amino-acid sites, 16 CPU... and I use 32 CPU).




Plus, for the 75 others, I had this erreur (not for the 15 first fasta, I don't understand) :





ERROR: STACK TRACE FOR DEBUGGING:
ERROR: 1   double* aligned_alloc<double>(unsigned long)
ERROR: 2   PhyloTree::initializeAllPartialLh(int&, int&, PhyloNode*, PhyloNode*)
ERROR: 3   PhyloTree::initializeAllPartialLh()
ERROR: 4   computeSiteFrequencyModel(Params&, Alignment*)
ERROR: 5   runPhyloAnalysis(Params&, Checkpoint*)
ERROR: 6   main()
ERROR: 7   __libc_start_main()
ERROR: Not enough memory, allocation of 59743927072 bytes failed (bad_alloc)



Do you think my parameters -nt 8 -mem 60G are good ?


This is because the first phase uses a lot of RAM to estimate .sitefreq file. The doc shows you how to avoid this. Or you can use -mem option, but set smaller limit, e.g. -mem 20GB.




According to you, using PMFS, what would be the most pertinent ? What we want is the equivalent of a C60, without having as much computing time.

Yes, hope that I have answered your questions,

Minh




Best regards




Richard GOUY


--
You received this message because you are subscribed to the Google Groups "IQ-TREE" group.
To unsubscribe from this group and stop receiving emails from it, send an email to iqtree+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/iqtree/957939b0-efd8-478b-8435-ebe6d93be27d%40googlegroups.com.

[Richard Benji]

OK thank you for everything, I think I'm good. I just removed the -mem option, for now it works... It's been running for 1 day (I already have guide tree from previous analyses).

Richard

Le lundi 13 avril 2020 06:37:55 UTC+2, Richard Benji a écrit :

[Minh Bui]

I see, that’s a bug with -mem option that somebody else already reported, but we haven’t had the time coming to that yet. But for now please ignore -mem option.

Thanks

Minh

-- 
You received this message because you are subscribed to the Google Groups "IQ-TREE" group.
To unsubscribe from this group and stop receiving emails from it, send an email to iqtree+un...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/iqtree/84d078a3-1989-458f-8f0c-acee370c8827%40googlegroups.com.