Hi George, hi Ben, thanks a lot for explaining! Yes, I'm well aware of multi-model refinement/building works and even myself was involved into one: Acta Cryst. (2007). D63, 597-610. "Interpretation of ensembles created by multiple iterative rebuilding of macromolecular models". The "proper" way would be to enable using multiple models in phenix.refine (those defined with MODEL-ENDMDL cards). Currently this is not possible, but allowing it is in our to-do list. There is a number of technical issues that we need to address first (and some of them are not entirely up to me to re-solve: ed. Ralf's PDB interpretation procedure). Yes, using altLoc identifiers solves the problem indeed, although the under-the-hood calculations become extremely inefficient, but again, you are right, it works (see remark below for potential issues!). I can go ahead and remove "max=4" limitation. One remark. Recently I went through the whole PDB and tried to re-compute the reported statistics (R-factors, for example) for all entries containing multiple models. You can do it using phenix.model_vs_data tool: phenix.model_vs_data model.pdb data.hkl So, the observation is: more models your PDB file contains, less reproducible the R-factors. The obvious reason for this is the precision filed for occupancy in PDB file, which is 1.00. This means if you have 16 models, then you report occupancy as 0.06 and NOT 1./16 = 0.0625. So the rounding errors are the issue here. What if you have 200 models? Pavel. On 9/15/09 10:58 AM, George Phillips wrote:
Pavel,
Ben and I are here now, so here is your response from both of us.
We are trying to do complete ensemble refinements like we used to do with CNS. 8 or even 16 (or more)copies of the whole protein tends to give the best R-free. See the article below.
We can recompile if you tell us what needs to be changed (or give us some clues where to look), but we need a lot more than four. Clearly this is not the normal use of this feature, but it works. We are getting drops in Rfree from your test examples in phenix even with four.
Levin et al. Structure, 15: 1040 (2007).
George N. Phillips, Jr., Ph.D. Professor of Biochemistry and of Computer Sciences University of Wisconsin-Madison 433 Babcock Dr. Madison, Wi 53706 Phone/FAX (608) 263-6142
On Sep 15, 2009, at 12:22 PM, Pavel Afonine wrote:
Hi Ben,
to allow so I will have to slightly change the code... I went through the whole PDB and did not find any item that has more than 3 or 4 conformers (at the moment of coding this). So that made my choice for that temporary limitation of max=4 conformers (putting aside a number of cases of abusing altlocs to mimic multiple models MODEL-ENDMDL). Unfortunately, nothing is so permanent as temporary, so we have 4 since that -:)
May I ask you: why you need to have more than 4 conformers? If it is really a bottleneck and stops you from doing something important right now, I can go ahead and fix it.
Pavel.
On 9/15/09 9:20 AM, Ben Mueller wrote:
I am a relatively new Phenix user and I am trying to see if it is possible to push the number of conformers beyond 4. I tried to do so, and I recieved the error message:
RuntimeError: Exceed maximum allowable number of conformers (=4).
Is there an easy (or difficult) way around this?
Thanks for your time,
Ben Mueller
Phillips Lab Department of Biochemistry University of Wisconsin - Madison ------------------------------------------------------------------------
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] mailto:[email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
------------------------------------------------------------------------
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb