Posts by Ingemar

1) Message boards : Number crunching : Problems with Rosetta version 5.93 (Message 51147)
Posted 3 Feb 2008 by Ingemar
Post:
The 2h4o**** jobs were of a very large protein with very complicated architecture so rosetta gets stuck a
lot during model generation. No more jobs of this variety will be sent out due to the problems you report.
2) Message boards : Number crunching : Problems with Rosetta version 5.93 (Message 50671)
Posted 14 Jan 2008 by Ingemar
Post:
argh im getting angry

from the last 9 WU's i had 7 errored out. 7!!!!
thats 77.77%
today 2 more WU's crashed, but i dont feel like posting links anymore, its always the same stuff, sin and cosin thats out of range, when are you guys going to fix this. or give me a reply.?


Hi Luuklag,

The overall error rates of the WU that are crashing for you are much lower than what you observe (around 2-5%). You may be unlucky, on the other hand they are caused by the the same problem (the cosine error) and not only for one type of WU so we need to fix that. We are looking into this problem to find the bug.
3) Message boards : Number crunching : Problems with Rosetta version 5.93 (Message 50473)
Posted 8 Jan 2008 by Ingemar
Post:
In this case it was only 2 days between the boinc and ralph update which I agree is on the short side. However, I just want to alert you to one issue: how long one tests the code on ralph before submitting to boinc depends on the nature of the update. In this case the latest two updates on ralph concerned only code in one scientific protocol. The rest of rosetta stayed the same and the bulk of the jobs sent out were running on identical code base as the previous boinc version. As for the new code, we had tested it on ralph and locally without problems so we were condfident it would not mess things up.

Still, I generally agree that more time between updates is better. I will leave more time the next time I make an update. But as I alluded to above, how much testing is necessary on ralph must be determined on a case to case basis. Also, compared to many other projects we are doing more code development and probably need to update more often.

As for the latest problems: They were mainly caused by high-memory jobs run with higher priority run over holiday season when the queue was not completely full. We have noted these problems and are working on solutions to avoid these problems in the future.

We hope for your continued support, thanks!
4) Message boards : Rosetta@home Science : Rosetta@home Active WorkUnit(s) Log (Message 50444)
Posted 7 Jan 2008 by Ingemar
Post:
In the previous post Rhiju described a simulation called FOLD-AND_DOCK where we are predicting the structure of protein complexes from sequence alone. In a cell few proteins carry out their function on their own. Most proteins are either forming complexes with other proteins a fraction of the time (and the structure of these is something we try to predict with protein-protein docking simulations) or form permanent complexes consisting of multiple chains. Predicting the structure of a protein complex from sequence(s) of the involved chains is something that has never been shown, mostly because of the enormous search space that has to be covered in such a simulation.

A very important class of protein complexes are those that contains multiple copies of the same type of chain, we call them homooligomeric. A vast majority of these homooligomeric proteins contains some type of internal symmetry. For example, the most abundant form of homooligomers are dimers (two identical chains) and most are symmetrical: if you rotate one chain 180 degrees you get the other partner in the complex.

The fact that they are symmetrical allow us to reduce the search problem (all the chain are internally identical and are related by a set of rotations and translations specified by the type of symmetry). This makes it feasible to try to predict the structure of symmetrical protein complexes from sequence. Its still a huge search problem, and thats why we need your help!

Right now we are trying to predict the structure of a protein with pdb code 1zpy. The size of this protein is actually around 900 residues and if we succeed it would make the largest protein structure predicted to atomic accuracy from sequence (About 8 times larger than anything predicted before). We
are not actually simulating the whole protein, we can make use of symmetry to reduce the problem, but its still a large simulation. So if you get a job with the 1zpy in the name be patient!

Why are we doing this? Homooligomeric proteins are a biologically very important class of proteins. They are forming virus capsids, proteins that cleaves DNA, hemoglobin that transport oxygen, actin involved in muscle contraction, channels that transport ions and secretion systems used by pathogenic bacteria just to give a few examples.
5) Message boards : Number crunching : Problems with Rosetta version 5.93 (Message 50335)
Posted 5 Jan 2008 by Ingemar
Post:
Please post problems and/or bugs with rosetta 5.93. Thanks for your
support!
6) Message boards : Number crunching : Rosetta Application Version Release Log (Message 50334)
Posted 5 Jan 2008 by Ingemar
Post:
Rosetta has been updated to 5.93. This version includes an improved treatment of dihedral
symmetry of proteins. This is a common, biologically important symmetry found among protein
homooligomers (a collection of proteins with the same sequence and often structure). Thanks for
supporting rosetta@home!
7) Message boards : Number crunching : Problems with version 5.90/5.91 (Message 50326)
Posted 4 Jan 2008 by Ingemar
Post:
The 1zpy jobs are really big and complex simulations (which are more likely to get stuck) and a lot of them was sent out over the holidays. We will update the executable soon to improve the performance of this and other similar simulations in the future.
8) Message boards : Number crunching : Problems with Rosetta version 5.89 (Message 49676)
Posted 13 Dec 2007 by Ingemar
Post:
Please post any problems with rosetta 5.89 here. Thanks!
9) Message boards : Number crunching : Problems with Rosetta version 5.81 (Message 48891)
Posted 20 Nov 2007 by Ingemar
Post:
MFR_SYMM_FOLD_AND_DOCK_RELAX_GB1_mutant_2286_18566 was crashing because the name of the return file was incorrectly specified. So the jobs ran correctly and produced an output but the client tries to transfer a file that does not exists.

The job has been removed from the queue and will not be sent out more. With that said I apologize to everyone that we did not track this down earlier. It got under the radar.

10) Message boards : Number crunching : Problems with Rosetta version 5.81 (Message 48402)
Posted 5 Nov 2007 by Ingemar
Post:
Please report problems with this version. Thanks!
11) Message boards : Number crunching : Problems with Rosetta version 5.80 (Message 46590)
Posted 19 Sep 2007 by Ingemar
Post:
We discovered that the result files of the now infamous capri were quite large which has caused some problems. We are looking into this issue now, but rest assured, no more of these jobs are being sent out. So things should be back to normal now that none of these jobs are in the queue.
12) Message boards : Number crunching : Problems with Rosetta version 5.80 (Message 46479)
Posted 17 Sep 2007 by Ingemar
Post:
A large fraction of the CAPRI-something jobs are failing. We are removing these jobs from the queue now and will not run more of those before we located the problem. Sorry for the inconvenience!
13) Message boards : Number crunching : Problems with Rosetta version 5.80 (Message 46241)
Posted 15 Sep 2007 by Ingemar
Post:
It appears that some of the Capri docking runs get stuck and gets terminated by the watchdog. The watchdog seems to do its job, the problem seem to be the simulations. This is the first time we do large scale tests on some new simulation modes and we will have to analyze why some runs get stuck/crashes. CAPRI ( Critical Assesment of Protein Interactions) is a competion where we try to predict the structure of protein-protein complexes. We have a deadline for submission of our models to this competion coming up soon and thats why you see so many Capri-something jobs. they will soon be out of the queue.

And yes we did increase the watchdog timeout.
14) Message boards : Number crunching : Problems with Rosetta version 5.80 (Message 46153)
Posted 13 Sep 2007 by Ingemar
Post:
Please report problems with this version. Thanks!
15) Message boards : Number crunching : Lots of decoys! (Message 41233)
Posted 21 May 2007 by Ingemar
Post:
Yes! These are my runs and they are as you point out very fast. Suspiciously fast one might think, but everything is as should be. Thanks for keeping an eye on this!
16) Message boards : Rosetta@home Science : Rosetta@home Active WorkUnit(s) Log (Message 38199)
Posted 24 Mar 2007 by Ingemar
Post:
Jobs with names DOCKING_*rhj* is running protein-protein docking on dimers where the individual monomers are related by symmetry. The structures are coming from ab-initio structure prediction. The structure of this protein could not be solved by standard techniques used for determining crystal structures. Although a crystal could be grown and experimental data looks good one of the procdures in crystal structure determination, phasing, could not succesfully be carried out starting from similar proteins previously solved. The idea here is to create models with ab-intio+docking that can be used as starting points for the phasing procedure. If this works this could be way of rescuing data from x-ray crystallography data from biologically important proteins that can not be converted to 3D-structures, which would a significant breakthrough.

Thanks for your help!

17) Message boards : Number crunching : Problems with rosetta 5.48 (Message 37502)
Posted 5 Mar 2007 by Ingemar
Post:
Hi all, the HINGE jobs you run are for the current round of CAPRI (Critical Assessment of PRediction of Interactions) in which a the predictors are given the structure of two binding partners and are challenged to predict the structure of the protein complex during 2-4 weeks. This round involves a prediction of a structure of a homodimer. We do not have structure of the protein monomer in this case but know that it has high sequence similarity with a couple of other proteins which can be used as a template. The protein has two domains connected by a loop which we think act as a hinge. So its is pretty challenging problem where we have to simulate dimer of 750 aminoacids together with conformational flexibility in the hinge region. So we need a lot of computer power and memory in order to simulate this system.

Next time we will give heads up when we run something like this...

Thanks for your support!
18) Message boards : Number crunching : Rosetta Application Version Release Log (Message 37316)
Posted 1 Mar 2007 by Ingemar
Post:
Boinc is updated to 5.48. We are including a special symmetric docking protocol and a fix to the problem of frequent watchdog terminations of docking workunits. Keep us posted on bugs here!
19) Message boards : Number crunching : Problems with rosetta 5.48 (Message 37315)
Posted 1 Mar 2007 by Ingemar
Post:
Please report here for problems you have observed with Rosetta version 5.48.






©2024 University of Washington
https://www.bakerlab.org