Bug reports

[Anonymous]

Hi Maria et al., thanks for making this available, looks really promising. How do you want us to submit bugs? I see that the Bugs tab on Sourceforge doesn't allow users to open tickets (or am I missing something?). Regards, Cymon.

[Marcelo S Zanetti]

Hi Cymon,

Please report bugs here. When confirmed I move them to the BUGS section.

Thank you!

Marcelo

[Anonymous]

Hi Marcelo,

OK, this analysis appears to have completed its calculations and then falls at the final hurdle without any output...

Running on CentOS release 5.4, GNU libc 2.5, GCC 4.1.2.

The build is OMP enabled but not BLAS/LAPACK

Here's (relevant bit of) the log:

                           ..................

oooooooooooooooooooooooo CURRENT SETTINGS oooooooooooooooooooooooo
..................

. Sequence filename: ygmitDNA_noStops.phy
. Number of taxa: 60
. Sequence length: 10489
. Data type: Codon
. Alphabet size: 61
. Genetic code: Bacterial, Archaeal and Plant ...
. Sequence format: Interleaved
. Number of data sets: 1
. Number of bootstrapped data sets: 0
. Compute aLRT: Yes (SH-like branch supports)
. Model type: MG+W+K+F
. Equilibrium frequencies model: Optimized CF3x4
. Prop. of invariable sites: 0.00
. Number of subst. rate categs: 1
. Number of dn/ds rate ratio categories: 6 - Discrete Gamma model (M5)
. Gamma distribution parameter: Estimated
. 'Middle' of each rate class: Mean
. Optimise tree topology: Yes
. Heuristic during tree search: Original
. Parameter optimization strategy: Multi-variate (BFGS)
. Tree topology search: Best of NNIs and SPRs
. Starting tree: BioNJ + JC69
. Add random input tree: No
. Optimise branch lengths: Yes
. Optimise subst. model params: Yes
. Matrix exponential: Eigenvalue problem
. Run ID: None
. Random seed: 805243175
. Code optimization: OpenMP
. Version: 1.00 201305.23

oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo

[...]

. (120653 sec) [ -300178.5014] [dn/ds ratio ]
. (120653 sec) [ -300178.5014] [Alpha and Beta ][0.68 1.86 ]
. (120653 sec) [ -300178.5014] [CF3x4 freqs. ]

. Checking for NNIs, optimizing five branches...

. best_lnL = -300178.496325, c_lnL = -300178.501354
. Err in file utilities.c at line 11530

. Type enter to exit.

EOF

... and the output remains unwritten:

-rw-r--r-- 1 cymon biouser 0 Jun 3 12:07 ygmitDNA_noStops.phy_codonphyml_stats.txt
-rw-r--r-- 1 cymon biouser 0 Jun 3 12:07 ygmitDNA_noStops.phy_codonphyml_tree.txt

Regards, Cymon

[Marcelo S Zanetti]

Hi Cymon,

can you add the data set here? (i.e. Add attachment button) ... if you do not see that button then just send it to me at mzanetti@ethz.ch. Thank you!

[Marcelo S Zanetti]

Hi Cymon,

I got your file

how did you execute this run? command line or menu? can you paste here the command line or the sequence of screens until you get the execution started?

thx

[Marcelo S Zanetti]

The bug is confirmed. The problem is in the -s BEST option ... finalization is broken ... moved to bugs.

Thank you!

[Anonymous]

Hi Marcelo,

This is more an enhancement request than a bug report, and concerns the building of the BLAS/LAPACK enabled version on RedHat and derived distributions...

Can you remove the requirement to link to a library called liblapack_atlas ? This library appears to be Debian (and derivatives) specific and I can't find a package or build option for an equivalent and similarly named library on RedHat-based distributions. libblas, liblapack, and libatas are all present/available, but not liblapack_atlas.

Regards, Cymon

[Marcelo S Zanetti]

Hi Cymon!

This is easy, just use the make_phyml build script provided in the source file ... it does not use the any of the extra libraries (i.e. BLAS/LAPACK OpenMP). Let me know if this works.

[Anonymous]

Hi Marcelo,

Sorry, that wasn't worded very well.

Presumably there is a worthwhile computational advantage to have the BLAS/LAPACK optimised versions. But at present, this appears only to be available to Debian-derived distributions due to this library dependency.

I was hoping you might remove the dependency on the Debian-specific library (liblapack-atlas) while still allowing the BLAS/LAPACK optimised version to be compiled on RedHat derived distributions. I assume the code you need from liblapack-atlas is available in differently labelled libraries on Redhat systems. (I am assuming: although the BLAS/LAPACK/ATLAS libs a quite a mess on RH.)

Cheers, C.

[Anonymous]

Hi Marcelo,

Sorry, maybe that was a bit ambiguous... I would like to build the BLAS/LAPACK optimised version on RedHat, but you have Debian specific dependencies...

Regards, Cymon

[Anonymous]

Hi Marcelo,

Ah, sorry, I was obviously not clear: I would like to build the BLAS/LAPACK enable version on a RH system (for the speed), but you currently rely on libraries specific to Debian (and derivatives).

Regards, Cymon

[Marcelo S Zanetti]

Ok, moved it to bugs and we will see what we can do.

Thx!

[Anonymous]

Hi Marcelo,

A few command line parsing things...

the "-w" d option appears not to work and defaults to "-w s"

"-w g" alone without specifying --wclasses defaults to "-w s" - should at least throw an error

"-wclasses" instead of "--wclasses" is silently ignored and "-w g" defaults to "-w s"

"--optimise" rather than "--optimize" is silently ignored

Regards, Cymon

[Marcelo S Zanetti]

Hi Cymon,
I am forwarding that to Stefan. We take care of that.
Thank you.

[Stefan]

Hi Cymon

Thank you for your comments. I changed some things in code release r285; maybe you want to recompile and see if it works better for you? Let us know if things are ok, we will then also recompile the binaries.

A detailed answer:

the "-w" d option appears not to work and defaults to "-w s"

"-w g" alone without specifying --wclasses defaults to "-w s" - should at least throw an error

In the previous versions, the number of omega classes defaults to 1. This is not documented in the manual or the help text (I'm sorry for this), but it makes the result understandable; using the 'd' or 'g' option with just one single omega class defaults back to a model with a single omega parameter, which is the same as the 's' option.

In release r285, I changed the default number of omega categories to 4 which is probably more common. I also have documented it now in the help page ('-h' switch). It's not included in the manual yet, but this will be done too.

"-wclasses" instead of "--wclasses" is silently ignored

"--optimise" rather than "--optimize" is silently ignored

I could not reproduce this behaviour on my machine but changed some parameters in the input parsing method. I hope this bug has been killed now; otherwise, I would be happy if you could send me some example input parameters to reproduce this.

Thanks again,
stefan.

[Marcelo S Zanetti]

There is a new source code package if you like

http://sourceforge.net/projects/codonphyml/files/codonPhyML_windows_binary_1.00_201306.10.exe/download

Best

[Anonymous]

Hi Marcelo and Stefan,

Can you confirm this observation...

CodonPhyML 1.00 201306.18

In the command options:

-w (or --omega) model (Required if a codon model is used)
The omega parameter or nonsynonymous/synonymous rate ratio.

    model = DM0   : (default) Single omega model.
    model = DMODEL: Discrete unconstrained distribution model.
    model = DGAMMA: Discrete gamma model.

If you don't specify -w on the command line the omega parameter defaults to a fixed value of 1.0. (at least in the log file it it reported as 1.0 and there appears not to be any recorded estimation of the parameter during the run.)

If you do specify the -w on the command line "-w s" the omega defaults to single estimated value.

In the manual (P 116 under --wvals) it says Per default, all values for w are estimated by ML. Even if the manual is inaccurate, I still wouldn't expect the model to differ depending on whether or not I specified a default value.

Best regards, Cymon

[Marcelo S Zanetti]

Hi Cymon, sorry for the delay. I will look into it and will let you know asap. Thank you!

[Anonymous]

I am running CodonPhyML on Mac OS through the binary file that I have not modified at all. In the PDF manual it says that I have to enable OpenMP- how do I do that on the Mac. I did not need to configure anything. I also wonder how you change the path to the Yaml (or is this not needed? I did not understand from the manual what the Yaml did for the program). I also wonder how I use the blas for the same reason as above (since I never had to configure anything).
Greta

[Marcelo S Zanetti]

Hi Greta,

Thank you for your interest.

The MAC binary does not support OpenMp neither the Blas library yet. I am not sure, but I think YAML is not needed anymore. Will confirm that with Stefan, and will write back to you.

I will take care of updating the manual. Thank you for pointing this out. If you need anything, please let me know and we will try fix it for you.

Best

Marcelo

[Anonymous]

Hi, first of all thanks for the software. I have run an analysis using the aLRT as support measure and the values that I have obtained on the branches are > 1. I guess these are the 2(L1-L2) values and not the P values. Is this a bug or is it correct? Also, if these are the ∆Likelihood values, do I use 1 as degree of freedom to obtain the critical value?

Thanks for your help

Ramiro

[Marcelo S Zanetti]

Hi Ramiro, sorry for the late reply. Sourceforge is not notifying me correctly on news posts here.

Will check and return an answer to you later.

Best

[Marcelo S Zanetti]

Hi Ramiro, this is the answer I got from Dr. Anisimova, she developed the aLRT method. Let me know if this helps.

Best

Marcelo

---

Hi Marcelo,

that's right it seems he got the aLRT statistics and they are not converted into branch supports (command option -b -1).
He simply has to choose the other option ( -b -2), it is called approximate likelihood ratio test with Chi2-based branch supports (or something like that, I don't remember on the menu).
But you can tell him that we now recommend using aBayes (-b -5) accoding to Anisimova et al 2011 (Syst Biol):
http://sysbio.oxfordjournals.org/content/60/5/685

Hope you are doing well,

Maria

PS: Here are the options we have for branch supports:

-b (or --bootstrap) int
    int >  0: int is the number of bootstrap replicates.
    int =  0: neither approximate likelihood ratio test nor bootstrap values are computed.
    int = -1: approximate likelihood ratio test returning aLRT statistics.
    int = -2: approximate likelihood ratio test returning Chi2-based parametric branch supports.
    int = -4: (default) SH-like branch supports alone.
    int = -5: approximate Bayes (aBayes) support.

---

[Anonymous]

Hi Marcelo,

Thanks for your reply, it is very helpful! I wasn't sure whether using a bayesian method to estimate branch support on a likelihood tree would be appropriate, but I'll use that method after Anisimova.

Best

Ramiro

[Marcelo S Zanetti]

Ok, don't hesitate to write if you have any further problems. Best!