Re: [phenixbb] low resolution refinement question

15 Oct 2009

...
- Could you give  some tips about refinement at this resolution other 
than rigid body refinement - Can TLS and grouped B-factors work at this 
poor resolution ?
I did the refinement of a low resolution (4.1 A) structure of a 20S 
proteasome where I used one TLS group per domain or per ring, and one 
B-factor per domain.  I used R-free set selected in shells because of 
the high degree of NCS but didn't get a significantly higher R-free than 
selecting at random (did not study this at great length).  Actually I 
think this structure was the reason I first started using phenix.refine, 
because AFAIK this is the only program that can do this combination (TLS 
and grouped B).

Context: 20S has 14 copies each of 'A' and 'B' subunits (segids AA* and 
BB* respectively).

Something like this (edited to spare you all 28 segids):

refine {
   strategy = *individual_sites rigid_body individual_adp *group_adp *tls
   adp {
       group = "segid AAA1"
       group = "segid AAA2"
       group = "segid AAA3"
       group = "segid AAB1"
       group = "segid AAB2"
         etc
       one_adp_group_per_residue = False
       tls = "segid AAA*"
       tls = "segid AAB*"
       tls = "segid BBA*"
       tls = "segid BBB*"
   }
   ncs {
     find_automatically = False
     restraint_group {
       reference = "segid AAA1"
       selection = "segid AAA2"
       selection = "segid AAA3"
       selection = "segid AAA4"
       selection = "segid AAA5"
       selection = "segid AAA6"
       selection = "segid AAA7"
       selection = "segid AAB2"
           etc
       coordinate_sigma = 0.07
       b_factor_weight = 10
     }
     restraint_group {
       reference = "segid BBA1"
       selection = "segid BBA2"
       selection = "segid BBA3"
       selection = "segid BBA4"
       selection = "segid BBA5"
       selection = "segid BBA6"
       selection = "segid BBA7"
       selection = "segid BBB2"
          etc
       coordinate_sigma = 0.07
       b_factor_weight = 10
     }
   }
}

Not exactly the best structure I've ever done, but 14-fold averaging was 
useful.  In the above case I was doing TLS on entire A7 or B7 rings 
(segid AAA* = segids AAA1-7) rather than per-domain TLS.

PDB code 3H4P
REMARK   3   R VALUE            (WORKING SET) : 0.254 

REMARK   3   FREE R VALUE                     : 0.325
...
- I would really prefer that rigid body refinement will strictly obey 
the NCS that was found in the molecular replacement solution. I
Why particularly trust the molecular replacement solution with an 
imperfect model ?  If you are just doing rigid body refinement, I'd 
argue that the rigid body results *define* the NCS at least for that 
model.  NCS restraints would have no effect on the results of rigid body 
(if the boundary definitions were consistent).

Phil Jeffrey
Princeton