Here is Randy's reply:
Dear Hari,
Well, it turns out to be a combination of some bad data and a
poorly-considered feature in Phaser.
Something must have gone wrong with the data from 2.9 to 2.8A, because
the intensities are much too large. I've been developing a new
relative Wilson plot tool to look at such things. This plots the
logarithm of the mean value of the observed F^2 divided by the mean
value of the calculated F^2 in each resolution shell, as a function of
(sin(theta)/lambda)^2. This is then scaled to an average curve from
something like 15000 data sets in the PDB. The plot in the attached
file compares the curve from your data (blue line) with the average
from the PDB (black line, with grey shading for the variation within
the PDB). Things go bad around 2.9A.
The single biggest reason that the map from Phaser is much worse than
maps from Refmac or SIGMAA is that Phaser assumes that the Fo values
are on the right scale; the Fc values in the high resolution shell end
up being scaled up, so you get very large map coefficients in this bad
resolution shell. If you look at the map in coot but limit the
resolution to 2.9A, it looks much better. (You can set resolution
limits in coot using the "Expert Mode" button in the open map from MTZ
window, but you have to give both low and high resolution limits.) In
Refmac and SIGMAA, the SigmaA values will refine to zero for the high
resolution shells, so that the map coefficients are zero in that shell
as well.
Maps from Refmac will probably look much better for a second reason:
your data are very incomplete beyond about 5A resolution. By default,
if you have hkl values in the MTZ file (e.g. if you've used uniquify),
Refmac fills in the missing data with D*Fc.
Apart from the highest resolution shell being severely downweighted,
the refinement of SigmaA values is probably relatively unimportant for
the rest of the data.
Thanks for bringing this up and sharing the data! We'll have to
figure out how to make Phaser cope with cases like this, either
through the SigmaA refinement or by looking at the relative Wilson
plot.
Regards,
Randy
On Fri, Feb 5, 2010 at 6:33 AM, Randy Read <[log in to unmask]> wrote:
> In all versions of Phaser up to the currently distributed ones, the assumed RMS error of the model is used to generate an a priori SigmaA curve for the likelihood target. At the end of the structure solution, the map coefficients are computed by using those a priori SigmaA values to calculate m and D.
>
> So it should indeed be better to go into SIGMAA, Refmac or phenix.refine and generate a map, because the SigmaA values will be refined to reflect the actual agreement between Fo and Fc, not the agreement that was guessed beforehand. Nonetheless, I'm surprised that the difference in quality can be so noticeable. That implies that the initial guess of model quality was far off the actual value, so it would be good to know what Phaser was told about model completeness and either expected RMS error or sequence identity, in cases where the maps were bad.
>
> We (by which I mean Airlie) are currently working on putting a SigmaA refinement into the end of the Phaser MR run, so future versions of Phaser won't have this issue. It would be useful if we could get an example or two of where the current version gives really bad maps, so we can verify that the new procedure fixes the problem.
>
> Regards,
>
> Randy Read
>
> On 5 Feb 2010, at 00:48, hari jayaram wrote:
>
>> Hi ,
>> I just switched to ccp4-6.1.3 and ran the phaser which is bundled
>> with the ccp4-6.1.3 release. After molecular replacement I get a
>> pretty good solution (TFZ 59.2) and output pdb and mtz files.
>> I am then looking at the maps from the phaser output mtz i.e the FWT ,
>> PHWT map and the FWT PHIC map . Both maps look very crappy and look
>> like noise. I tried the fft inside of coot and separately as a FFT
>> inside ccp4 and these maps look bad. So the FWT PHWT calc seem to be
>> off.
>>
>> However , if I take the phaser solution output model and do the sigmaa
>> myself. The map looks normal and sensible.
>>
>> It seems like phaser inside of ccp4-6.1.3 is not generating the output
>> mtz correctly .
>> Anyone else seeing this or is there something wrong with my setup.
>>
>>
>> Hari
>
> ------
> Randy J. Read
> Department of Haematology, University of Cambridge
> Cambridge Institute for Medical Research Tel: + 44 1223 336500
> Wellcome Trust/MRC Building Fax: + 44 1223 336827
> Hills Road E-mail: [log in to unmask]
> Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk
>
>
|