[phenixbb] handling of alternate conformations

Pavel Afonine PAfonine at lbl.gov
Mon Jul 16 10:50:03 PDT 2007

Hi Ulrich,

One thing I'm puzzled about (before reading your email until the end)... 
Normally we use NCS restraints at lower resolutions and availability of 
high resolution data may suggest to turn NCS restrains off completely 
(at least it's a definitely good thing to try). The alternative 
conformations are normally visible at high resolution (exactly where the 
NCS restrains are most likely useless if not harmful). So, it's 
interesting how you arrived at the situation when you have both:  need 
to use NCS restraints and alternative conformations.

> we have a complex with 4 chemically different polypeptide chains in the  
> asymmetric unit, composition A3B3CD. For chains C and D we see additional  
> density, which appears to belong to three alternating conformations which  
> can be thought as rigid-body motions (for the whole chains C and D, let's  
> call these conformers 1,2 and 3, but we label them conventionally in the  
> file with the prefix A,B,and C) with roughly estimated occupancies of  
> about 0.3, 0.3 and 0.4 (we can really see most of them). A and B have no  
> visible alternating conformations (occupancies set to 1.0). From the atom  
> count it appears that phenix reads in three conformers called A,B and C  
> with A: A3B3C(1)D(1), B: A3B3C(2)D(2) and C: A3B3C(3)D(3).

OK, now I see. Did you try TLS to model this? Not sure if it's best 
thing to do if you actually see the density for alternative 
conformations, but just out of curiosity...

> Phenix.refine says it has these 3 conformers:
> conformer A A:2700 atoms,
> conformer B:2700 atoms, common with A 2100,
> conformer C:2700 atoms, common with A 2100, common with B 2100.
> Refinement yields an R/Rfree of about 0.210/0.325.

Yes, the gap between R and Rfree looks not good (too big).

> If we treat the three conformers of polypeptide chains C and D as  
> independent chains with occupancy each 1.0 (this is possible 'cos there  
> are no repulsive contacts between them, fortunately), we get R/Rfree of  
> 0.243/0.291 and phenix applies NCS restraints to all chains.
> The resolution is about 2.9 A. We use phenix.refine from the CCI_APPS,  
> version 2007-03-26-1903, SUSE Linux.
> My questions is: is the difference due to the lack of NCS between  
> conformers or are the occupancies mistreated somwhow? I am asking because  
> I would not consider the whole A3B3C(x)D(x) particle as a conformer but  
> only chains C and D ......
> Is there a possibility to apply NCS restraints to conformers?

- If you do:
% phenix.refine model.pdb data.hkl main.ncs=true
phenix.refine tries to automatically determine NCS groups and use them 
in NCS restraints. This may be not good in your very specific case.
- If you do:
% phenix.refine model.pdb data.hkl main.ncs=true ncs_groups_selelctions.txt
where ncs_groups_selelctions.txt is the file where you provide 
selections for NCS groups manually, then phenix.refine will use them but 
it will try to optimize them automatically. This is also may be not good 
in this specific case.
- Most likely you need to specify your own selections for NCS groups but 
disable their automatic correction:
% phenix.refine model.pdb data.hkl main.ncs=true 
ncs_groups_selelctions.txt  ncs.find_automatically=False
Look "Atom selection examples" paragraph in phenix.refine Documentation 
to see the syntax for selections involving alternative conformations.

2) Once you figured out how you deal with NCS selections, add to your 
normal refinement run group occupancy refinement, so on top of 
coordinates and B-factors refinement you also refine one occupancy 
factor per selected chain (that you believe in alternative conformation):

% phenix.refine model.pdb data.hkl main.ncs=true 
ncs_groups_selelctions.txt  ncs.find_automatically=False 
refine.occupancies.group="chain A and resid 1-50" 
refine.occupancies.group="chain B and resid 1-50"

The command above will do three macro-cycles of NCS restrained xyz and 
b-factors refinement, plus group occupancy refinement: one q per chain A 
and one q per chain B (all other non-selected occupancies will be fixed 
to original input values).

Please not that currently we have no specific constraints on occupancies.

I hope Ralf may have more comments on this (he implemented NCS restraints).


PS> I will be away (not available by email) from Tuesday 17 until Monday 23.

More information about the phenixbb mailing list