Problems with DAMAVER versions

Calculation of SAXS and SANS profiles (CRYSOL, CRYSON), superposition of models (SUPCOMB, DAMAVER, DAMCLUST), database DARA
Post Reply
Message
Author
bharri17
Member
Posts: 3
Joined: 2013.09.17 18:49

Problems with DAMAVER versions

#1 Post by bharri17 » 2013.09.17 19:08

I have been utilizing the ATSAS suite of programs to compare a computational model with small-angle neutron scattering data. I have run into a problem with DAMAVER where a previous version of the software utilized by a neutron scientist gives a vastly different output compared to what I obtain using the latest version on my laptop, even when starting with the same GNOM output file. Our approach is as follows:

(1) Run DAMMIF 50 times supplying the GNOM output file, from the command line as follows:
for %i in (1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50) do dammif --prefix=DAMMIF_run%i --mode=slow --symmetry=P3 GNOM_outputfile.out
(2) Delete the -0.pdb solvent files
(3) Run DAMAVER on the 50 -1.pdb files to obtain an average structure.
- I do this on my version by running the following command: damaver -a -s P3 *-1.pdb (which runs damaver on all the -1.pdb DAMMIF files)
- The staff scientist accomplishes this using the following command: damaver /a P3 (which runs damaver on all .pdb files in the folder)

Can someone explain why we are getting such vastly different results using the same commands, with only the damaver version being different? I have attached the GNOM output file which is used for DAMMIF, as well as the average structure obtained from DAMAVER for each of us.
Attachments
damaver_newversion.pdb
(65.82 KiB) Downloaded 234 times
damaver_oldversion.pdb
(184.3 KiB) Downloaded 233 times
GNOM_outputfile.out
(33.67 KiB) Downloaded 211 times

franke
Administrator
Posts: 408
Joined: 2007.08.10 11:09
Contact:

Re: Problems with DAMAVER versions

#2 Post by franke » 2013.09.17 20:46

bharri17 wrote:(1) Run DAMMIF 50 times supplying the GNOM output file, from the command line as follows:
for %i in (1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50) do dammif --prefix=DAMMIF_run%i --mode=slow --symmetry=P3 GNOM_outputfile.out
This looks like c-shell? I only use bash, so in bash syntax this can be done as:

Code: Select all

for i in `seq -w 1 50` ; do dammif --prefix=run-$i --mode=slow --symmetry=P3 GNOM_outputfile.out ; done
bharri17 wrote:(2) Delete the -0.pdb solvent files
If you don't want it, try: "--omit-solvent".
bharri17 wrote:(3) Run DAMAVER on the 50 -1.pdb files to obtain an average structure.
- I do this on my version by running the following command: damaver -a -s P3 *-1.pdb (which runs damaver on all the -1.pdb DAMMIF files)
- The staff scientist accomplishes this using the following command: damaver /a P3 (which runs damaver on all .pdb files in the folder)
This runs damaver on the same set of *-1.pdb? Or did you run dammif 100 times, 50 times each and came up with different results?
bharri17 wrote:Can someone explain why we are getting such vastly different results using the same commands, with only the damaver version being different? I have attached the GNOM output file which is used for DAMMIF, as well as the average structure obtained from DAMAVER for each of us.
There have been quite some changes in damaver and friends and there's a good chance that you found a bug. Would you mind to send the 50 input files to my personal mail?

SaxsMax
Administrator
Posts: 349
Joined: 2007.08.03 19:16

Re: Problems with DAMAVER versions

#3 Post by SaxsMax » 2013.09.18 10:33

franke wrote: This runs damaver on the same set of *-1.pdb? Or did you run dammif 100 times, 50 times each and came up with different results?
Judging from the headers of the two damaver files,
the sets of the models used for averaging were different
P100D240xx-1.pdb and P100D240_xxrun-1.pdb
This might explain the vast differences in generated damaver files

bharri17
Member
Posts: 3
Joined: 2013.09.17 18:49

Re: Problems with DAMAVER versions

#4 Post by bharri17 » 2013.09.23 21:53

I apologize for posting to an irrelevant section.

As to your previous remark, the two sets of 50 dammif runs used to obtain the "old version" and "new version" damaver output files were indeed different. As I said before, they were run by two different people possessing two different versions of the ATSAS software package. However, they were generated using the same run conditions for both dammif and damaver, and the results are vastly different. Are you telling me that if I were to run dammif and damaver on a given out file using a given set of conditions, and repeat the process twice, I should expect to get vastly different results from run to run? I don't see how that makes any sense.

I will gladly send you my 50 input files to your personal e-mail as soon as possible. Thank you for your help.

franke
Administrator
Posts: 408
Joined: 2007.08.10 11:09
Contact:

Re: Problems with DAMAVER versions

#5 Post by franke » 2013.09.23 22:45

bharri17 wrote:However, they were generated using the same run conditions for both dammif and damaver, and the results are vastly different. Are you telling me that if I were to run dammif and damaver on a given out file using a given set of conditions, and repeat the process twice, I should expect to get vastly different results from run to run? I don't see how that makes any sense.
If this were true, why do you expect to get different models in your 50 runs to begin with? If all were determined you should certainly expect 50 identical models?

You may want to read a bit about Pseudo Random Number generation and the associated issues with random seeds. Maybe even the related section in the manual?

SaxsMax
Administrator
Posts: 349
Joined: 2007.08.03 19:16

Re: Problems with DAMAVER versions

#6 Post by SaxsMax » 2013.09.24 09:23

I think what is meant is that starting from one *.out file
the resulting averaged models should be similar (not identical).
But the results are vastly different

bharri17
Member
Posts: 3
Joined: 2013.09.17 18:49

Re: Problems with DAMAVER versions

#7 Post by bharri17 » 2013.09.24 18:14

Yes, this is exactly what I mean. Should I upload my 50 input files to damaver for you to take a look at? If so, how and where? Thanks again for your help.

franke
Administrator
Posts: 408
Joined: 2007.08.10 11:09
Contact:

Re: Problems with DAMAVER versions

#8 Post by franke » 2013.09.25 16:21

If you could send me one file of your two runs to "my-forum-name AT embl-hamburg DOT de" each to begin with?

Post Reply