[phenixbb] give-up-quick Phaser in Rosetta
Terwilliger, Thomas C
terwilliger at lanl.gov
Tue Jul 9 09:50:13 PDT 2013
Hi Morten,
Yes, this is a disadvantage of the way mr_rosetta runs for now. It tries everything, then picks the best. I can think of two options:
1. Try out phenix.MRage which is optimized to try many things in parallel in MR. Then take the best one and run mr_rosetta with it.
2. Run separate mr_rosetta jobs for each search model, and when you see one that is working, trash the others.
All the best,
Tom T
On Jul 9, 2013, at 10:41 AM, Morten Grøftehauge wrote:
Dear Phenixbb,
I was wondering if there was any way of getting Rosetta to give up easily and move on to another search model?
Say I have 20 or more equally awful search models. I think there's a possibility that one of them will knock it out of the park (i.e. TFZ > 8). But even if I have Phaser set to fast search I will end up with an exhaustive search because the search models are so bad.
fast_search_mode= True Run phaser with selection_criteria_rot_value and
then if no obvious solution, repeat with cutoff
lowered by search_down_percent
search_down_percent= 25 Used if fast_search_mode=True. Run phaser with
selection_criteria_rot_value and then if no obvious
solution, repeat with cutoff lowered by search_down_percent
If I set search_down_percent to 75 does this mean that Phaser will give up easily and move on to the next model? Is this a terrible strategy CPU-time-wise?
I am of course aware that search_down_percent is much more likely to give me the right solution for a given model but in this case there are just so many possible search models.
Cheers,
Morten
--
Morten K Grøftehauge, PhD
Pohl Group
Durham University
_______________________________________________
phenixbb mailing list
phenixbb at phenix-online.org<mailto:phenixbb at phenix-online.org>
http://phenix-online.org/mailman/listinfo/phenixbb
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20130709/bfc5a44f/attachment.htm>
More information about the phenixbb
mailing list