[phenixbb] problems with map quality, refinement and R/Rfree

[Joe Smith]

Hi all,

We have data sets for a protein-RNA complex at 3.2 A resolution. . The
data belong to the space group I4122 and contains one molecule of
Protein-RNA complex (300AA and 16nt; 66% solvent content) in the asym
unit. We have managed to get phases using Se-SAD and the present model
contains 60% of the protein atoms (250 residues many of them build as
Ala) and almost complete RNA (16nt).  We could not locate ~20 % of the
residues (present in various loops as well as at N- and C-terminal
ends). First 120 residues present at N-terminal end seem to have poor
relatively main-chain density and almost no side chain density. RNA
and 130 residues present at the C-terminal end have good electron
density for main-chain as well as side chain.

The solution seems to be correct as molecular replacement trials with
this model gave same solution in Phaser, Molrep and AMoRe. Solution
seems to be also correct because as expected one RNA strand present in
asym unit pairs with another one coming from symmetry related
molecule.

We have checked data in space group I4122 with CCP4 (SFCHECK,
cumulative intensity distribution), Yeates server as well as
Phenix.xtriage. In all the cases it indicates no twinning. However if
I process the data in lower symmetry space group it does indicate
presence of almost perfect twin. We realize that this could be just
because of processing data in lower symmetry space group and data in
all probability may still be fine.

The problem is now with refinement and model building. Refinement with
CNS and Refamc gave considerably higher value of R and R free. In
CNS1.1, refinement with small changes in the model some time leads to
large shift in R and R-free value.
 Refmac: R=36; Rfree=47
CNS1.1:  R=40-59; R-free=45-63
However in both the cases map looks more or less same (with reasonably
good density for the main-chain as well as RNA)

I tried refining with phenix.refine and there is some improvement in
the R(34%) and Rfree(40%) values. I think may be robust bulk solvent
correction incorporated in phenix.refine has helped in this case.
However, I still see no improvement in the map quality for the first
120 residues.

In the absence of any clear density I am unable to build any further.
I feel N-terminal domain is bit flexible and may have overall poor
density. I have used Se position as well as predicted secondary
structure in assigning the amino acid in the map but due to few breaks
in the N-terminal domain as well as poor density I am unable to assign
any amino acid into the poly ala main chain.

Frankly speaking, I do not know how to proceed further. I welcome any
kind of suggestions which could help us in this case.

Regards
Joe