Re: [modeller_usage] high model clashscore

[Thomas Evangelidis <>]

You must perform more though homology modeling. These are the parameters you can adjust:
            a = automodel(env, ......)
            a.starting_model = 1
            a.ending_model = 100
            # Very thorough VTFM optimization:
            a.library_schedule = autosched.slow
            a.max_var_iterations = 300
            # Very thorough MD optimization:
            a.md_level = refine.very_slow


and for loop:

            a.loop.starting_model = 1           # First loop model
            a.loop.ending_model   = 10         # Last loop model
            a.loop.md_level       = refine.very_slow # Loop model refinement level

I've taken this section from my script, of course you can use different values. Also have a look at dopehr_loopmodel class for automatic loop modeling.


If you are looking for a more drastic solution, then Energy Minimization is the answer. PyRosetta is my favourite. ClassicRelax protocol with the 'standard' score function in conjunction with the 'score12' patch ('score12' patch is not included in the example script) that incorporates terms for ramachandran and omega angles will improve your clashscore whilst reduce the ramachandran outliers, bad bonds and angles. All-atom protein relaxation script with instruction can be download from http://pyrosetta.org/scripts.html. 

Otherwise, if you prefer a less complicated yet descent energy minimization utility, I would recommend UCSF Chimera. You can do energy minimization from Tools->Structure Editing->Minimize Structure. Chimera is faster and also allows you to keep an atom selection rigid, for instance the active site residues (an advantage over PyRosetta). 

Alternative options include Molecular Dynamics or Monte Carlo tools. As 
far as I am concerned, I've tried NAMD and GROMACS in the past, but 
although they are efficient in reducing steric clashes, they both 
increase ramachandran outliers, bad bonds and angles, as judged by 
Molprobity server.

Additional improvement can been done by flipping side-chains to adopt 
statistically favourable conformations. The latter can been done with 
SCWRL4 and is recommended for homology models derived from low sequence 
similarity templates and before energy minimization. Great care should 
also be taken to preserve side-chain conformation of biologically 
important residues (i.e. in the active site of an enzyme). SCWRL4 will 
probably flip them too, which is something you wouldn't want. You can 
correct that by editing the .pdb file upon processing with SCWRL4. You can also do side chain optimization with the Chimera GUI.

hope this helps,

Thomas

On 20 November 2010 13:03, Anton Iershov <" target="_blank">> wrote:

Dear Modeller Caretaker,

When performing homology modeling with default parameters ("model-default.py") and subsequent analysis with Molprobity (http://molprobity.biochem.duke.edu/) I obtain models with much more high clashscore than the template has, and model optimization does not help much.

Then, Kinemage (http://kinemage.biochem.duke.edu) shows a lot of "bad" van der Waals contacts.

What should I do to obtain model of higher quality (I mean to reduce clashscore and number of "bad" contacts)?

Thanks in advance,

Anton.

_______________________________________________
modeller_usage mailing list
" target="_blank">
https://salilab.org/mailman/listinfo/modeller_usage