Re: [modeller_usage] strategies for model refinement
To:
Subject: Re: [modeller_usage] strategies for model refinement
From: Thomas Evangelidis <>
Date: Thu, 6 Jan 2011 04:40:34 +0200
I am aware that PyRosetta doesn't scale well with chain length but I'm not sure what you mean by "isn't useful for proteins with >200 residues". Is this the message in RosettaCommons forum where you got this information?
I use PyRosetta to relax proteins up to 850 aa long, it takes 5-6 days for each job on a 1.8GB processor, but the results are still impressive. IMO a good model wrt torsion angles, bond lengths and steric clashes is worth the time. For your protein length I would run 10 jobs and select the best model among them. In my experience more that 10 would be needless. Otherwise you can use an MD package which is much faster, but don't expect to be thrilled with the results.
With respect to loop refinement, this is the section that described the default optimization and refinement protocol:
I personally prefer to run loop modelling after model building so that I can keep the rest of the protein rigid, apply secondary structure restraints, distance restrains, etc. You can find an example script under modeller9v8/examples/automodel/loop.py that does loop modeling from an initial conformation.
This is what I had to suggest considering my 1 year experience with MODELLER and structure prediction in general. If anyone else disagrees or wishes to add something to my hints, please feel free to do so.
Thomas
On 6 January 2011 03:24, Irene Newhouse <">> wrote:
I'm a relatively new user of modeller & successfully built a fairly nice model of a protein with 258 residues. It forms a homodimer, and has to be modeled as a dimer, because there's a very flexible loop that can easily adopt conformation that prevent dimerization if the 2nd unit isn't present. [I confirmed that by trying it].
Someone on this group suggested the Molprobity server as a way to assess model quality, for which I'm very grateful. He also suggested relaxing the structure with Rosetta, & after epic battles with python setup & my operating system, I got it up & running. Only to learn from the RosettaCommons forum that Rosetta relax doesn't scale well with chain length & isn't useful for proteins with >200 residues. I managed to get the Molprobity score down to a respectable 2.-something using lots & lots of iterations of Schrodinger Prime loop refinement & side chain minimization. It took about a week of setting up a loop, running it, checking to see if it refined, iterating 'til I could move on, etc. So I'm looking for something a bit less hands-on, if possible. Also, Schrodinger is expensive & my boss may not renew the licenses when this lot expires in March. And finally, Molprobity signaled several C-beta issues in my structure, & Schrodinger just doesn't seem to do much about those. Therefore, I'm looking for other suggestions, preferably freeware, for model refinement. I have tried some loop refinement with Modeller itself, but wasn't thrilled with the outcome. This could easily be because I couldn't find sample scripts to copy, or an explanation of how to tweak the parameters if the loop doesn't seem to be refining. If anyone can point me in those directions, I'd be grateful, too.