[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: RE: RE: Segmentation Fault during modelling with modeller6v



Hi,

The problem with that one .ali file is the nature of the loops you
select/create. You have to really scrutinize that alignment, making sure
that one-residue gap insertion there is what really makes that alignment
better then some other possible arrangement.

As for the random seed: I don't think it makes any difference. Andrej
might be able to add something about this?

Thanks,
Bozidar

-----Original Message-----
From: 
[">mailto:] 
Sent: Wednesday, April 24, 2002 3:56 AM
To: 
Subject: re: RE: RE: Segmentation Fault during modelling with modeller6v


Hi,

i started the  job with SET GAP_EXTENSION = 0 0 yesterday with both
alignments 
and one of the jobs has crashed againat the 100th model as well, the
other job 
is still running 450+models. The top and pir filesof the job that
crashed  are 
attached.

I will start 10 jobs with each generating 125 models. To see if in every
run 
the 100th model crashes. As soon as i have results I let you know. One 
question about the seed, is it wise to take this randomly between -50000
and 
-2 like -44656,  -35435, -9835, - 15684, -378 ... or is it oke to take
in 
order like  -32768, - 16384, - 8192, -4096, -2048...

Thanks in advance

Kees-Jan Francoijs


- - - - - - - - - - - - - - Original Message - - - - - - - - - - - - - -
On Tue Apr 23 18:03:13 2002, "Bozidar" <>
wrote:
>
>Hi,
>
>These are the observations I got so far:
>
>1. Your OK job is OK with me also. It's still running (200+ models).
>
>2. Your crash job crashed at me as well at 99th model. What happened is

>that energy of the 100th model (which never got written out) went 
>through the roof (NaN) and the system never recovered from it.
>
>The workaround is as follows: you can run this job in parallel. Instead

>of generating 1000 models in one run, you can start maybe 5 jobs that 
>will generate 200 models each. All you need to do it to give a random 
>seed number in the top files (between -50000 to -2) to each of them: 
>SET RAND_SEED = -xvz
>
>Why this particular alignment crashes and not the other one is hard to 
>say. As I mentioned in my earlier email, you are selecting 50% of your 
>protein as loop region. Given the close similarity between two 
>alignments, it might be that 5 vs. 4 loops results in formation of bad 
>structure at some point (knots maybe), that simply collapses the 
>system.
>
>
>Al least what we know is that MODELLER reports no errors and that it 
>runs fine. I think that you have very specific case here, which will 
>require careful modeling on your part, making sure you control each 
>step of the way.
>
>Still, please let me know how happens with SET GAP_EXTENSION = 0 0
>
>Happy modeling,
>Bozidar
>
>
>
>-----Original Message-----
>From:  
>[">mailto:]
>Sent: Monday, April 22, 2002 3:52 AM
>To: 
>Subject: re: RE: Segmentation Fault during modelling with modeller6v1
>
>
>Hi,
>
>Sorry for the delay. I send you the requested files. (ed)
>
>When running this top pir and pdb file the job crashes (crash.zip). 
>When using another pir same top and pdb file everything goes 
>oke(oke.zip). The log file a
>is  included as zip file AVR4.zip in crash.zip. If there is anything
>else you 
>want to know just ask.
>
>with regards
>
>Kees-Jan Francoijs
>- - - - - - - - - - - - - - Original Message - - - - - - - - - - - - - 
>- On Thu Apr 18 19:06:31 2002, "Bozidar Jerkovic" 
><> wrote:
>>
>>Hi,
>>
>>1. Please send me complete set of input files plus log files and I'll
>>run it on 6v2 (not public yet) to see if it happens there.
>>
>>2. Yes we went through this, as well. Prosa is not sensitive assessing
>>loops. Please read the comment on the use of statistical profile for 
>>assessing loops, in Andras Fiser's Prot Sci 2000 loop modeling paper. 
>>People should use modeller scores to rank them.
>>
>>Thanks,
>>Bozidar
>>
>>
>>-----Original Message-----
>>From: 
>>[">mailto:]
>>Sent: Tuesday, April 16, 2002 8:00 AM
>>To: 
>>Subject: Segmentation Fault during modelling with modeller6v1
>>
>>
>>Hi Bozidar,
>>
>>I have been using modeller for 2 years now . I'm using  modeller since
>>version 4. Recently I ran into a few problems maybe you can point me
in
>
>>the right
>>direction.
>>
>>Two Questions:
>>
>>1.)  When running modeller 6V1 on a pentium III redhat 7.2 with all 
>>recommended patches applied. The following problem occurred. After  I 
>>started a job doing loop modelling i generated 1000 models and every 
>>thing went fine.
>>I changed the alignment a bit and try to run for an other 1000 models
>>the job 
>>crashes after 198 models. So i decided to run on a other machine,
>>different 
>>one 1.7 Ghz Dual Xeon,  same problem. What can be the source of the
>>problem. 
>>The job crashes with segmentation fault error.
>>
>>2.) I assessed the 1000 generated models with ProsaII to find  the 20
>>best models. I also used the value from the objective function from 
>>modeller to
>>rank them. When doing this it was easy to see if there is any
>>correlation 
>>between the rankings and guess what,  there is none actual it seems
>>random. Is 
>>there an explanation for this and which one is right or did i do
>>something 
>>wrong ? This surprised me since Andrej uses the ProsaII score in many
>of
>>his
>>articles as an objective criterion for the quality of the models and
>one
>>would expect the objective function of Modeller to be related in 
>>quality.
>>
>>
>>with regards
>>
>>C.J.J Francoijs
>>
>>
>
>- - - - - - - - - - - - End of Original Message - - - - - - - - - - - -
>
>
>

- - - - - - - - - - - - End of Original Message - - - - - - - - - - - -