Parrot NCS averaging

From: Peter Canning
Date: 5 October 2011 15:42

Dear CCP4'ers,

I am working on refining the structure of a protein complex consisting of two different chains. Data were collected to 2.3A in space group C2 and phased by molecular replacement without any problem, but the ASU contains 6 complexes (so 12 chains in total). In the ASU, the 6 complexes form 3 dimers. Dimer 1 is related to dimer 2 by a rotation operation and to dimer 3 by translation – I hope that makes sense!

When I run Parrot to do density improvement and NCS averaging, Parrot works beautifully (final FOM is 0.87) but NCS averaging causes the average NCS correlation coefficient to drop (from 0.94 to 0.64) and the average mask volume to increase from 0.31 to 0.7 (minimum volume increases from 0.05 to 0.07 and maximum volume goes from 1.2 to 2), which seems a bit high.

Unless I have misunderstood something, I thought the NCS correlation coefficient should increase during averaging and the average mask volume should decrease. If this is the case, has anyone any idea what I'm doing wrong?

Also, at the risk of asking a dumb question, is there a particular way my Parrot modified file should be input into Refmac to get the best results? The FOM from my refmac runs never seems to be as good as that output by Parrot.

Thanks in advance, all advice gratefully received, etc etc…

Peter Canning

SGC

----------
From: <Herman.Schreuder

Dear CCP4'ers,

I do have a similar case with 8 molecules in the asymmetric unit, related by improper symmetry and the structure is solved by molecular replacement. I do have 2 questions:

1) Does it make sense to average to improve phases? The molecular replacement phases perfectly obey the non-crystallographic symmetry?

2) For refinement I do not need average maps since Buster and other refinement programs are very capable to handle ncs. However, I am not a computer and for manual rebuilding and display purposes it would be very good to use averaged maps. Which program could I best use for this: Parrot, DM, Solomon, other? None of them seem to be really made for molecular replacement phases. 

Also my many thanks for any help, advice etc.

Herman Schreuder

----------
From: Savvas Savvides

Dear Herman

In some cases density modification starting with MR phases can indeed prove very powerful. 

We recently used Parrot to bootstrap a difficult case at 4.3 angs and were able to obtain very good maps for entire missing domains. 

See Supplementary Fig 1 in:Verstraete et al 2011 BLOOD 118:60-68

and more technical info and fig5 in:

Verstraete et al 2011 Acta F67:325-331. 

The "Ncs mask radius" card in parrot proved to be an important one to experiment with, and is probably more important to do so at medium-low resolution. We provide details of our observations/implementation on page 330 of the actaF paper.

We also found that post-MR HL coefficients written out by Phaser were also key. So in subsequent runs, after having built and refined partial

models, we would do a phase calculation run in Phaser to obtain HL by 'fixing' the partial model via a template .sol file with 0 values for rotation/translation across the board. 

Best wishes

Savvas

----------
From: Kevin Cowtan

Not impossible if the NCS is along a special direction. I've rewritten the logfile stats in the latest version to give you a more meaningful and less confusing number. That's true for experimental phasing. When you start from MR, your initial phases contain the NCS implicitly. The structure factor magnitudes in this case have something to say about the differences between the molecules. Again, I think the stats in the new version may be more informative. FOM from density modification is overestimated! From refmac less so!

K