Help! weird thing

From: xiaoyazi2008
Date: 11 March 2012 07:33

Hi All,

I have an interesting thing to share. 

2.3A dataset with good quality, P21

Partial model is available (~60% of the target protein). 

It seems that there are 4 copies in the ASU (Matthews_coef 2.6, 53%solvent)

Molecular replacement gave two copies of the model (Z scores are R6.2, T6.2, R6.8, T13.4). The solution is very clear. It could not locate the rest two copies.

However, a quick refmac5 refinement gave a very high R factor. 

The funny part is the symmetry operation in Coot. 

As shown in the JPEG figure, it looks like there should be another two copies (based on strong fo-fc green map), which locate in the empty space between models found by Phaser. 

Why is that Phaser could not find the remaining two copies even there are strong fo-fc density?

Any suggestions...

Thanks a lot!

Zhihong

----------
From: Dale Tronrud


  It could be that your partial model has a loop, not present
in the true solution, that is causing a clash.  You could run
Phaser again with the anti-bumping restraint weakened or disabled,
or more carefully edit your partial model.

Dale Tronrud

----------
From: Garib N Murshudov

Hi

Could you please check:

1) If there is psedotranslation. It could be done by using sfcheck, molrep, ctruncate or calculating patterson map and displaying using coot at 8-10 sigma level (it is my favourite method for analysis of pseudo translations), whole unit cell ( a bit bigger than whole unit cell). Then if you see large no origin peak (very likely along one of the axis, could be a). If yes then you have several options: using phaser - 1) reduce cell, find solution in smaller cell and then expand; 2) use molrep to solve. When there are two copies related with pseudo translation molrep can give you solution; 3) as far as I am aware latest version of phaser works with pseudo translation. If you have pseudtranslation you should be aware that even if you solve the structure starting R factors could be 70-80%. Then you may want to do 40 cycles of rigid body and 40-100 cycles of ljelly body

2) Check your space group in pdb and mtz file. They may not be consistent.

I hope it helps.

Garib

<weird thing.jpg>

----------
From: Anastassis Perrakis

Hi -

I agree with Garib that its likely a pseudo-translation issue.

I also agree with that the advice he gives is correct, but ...

... since I am evidently less smart to follow all these steps, 

I like to use phenix.xtriage that will  tell me if there is pseudo-translation or not, 

and will give a p-value for that being significant. Its at the end of the text output.

I am not sure if Phaser deals these days with pseudo-translation - I guess Randy can tell us.

If not, there is a very simple trick to make Phaser work with pseudo-translation,

but since I threw the ball to Randy's court and he told me the trick a few years ago,

I will let him explain only if needed ;-)

Best,

Tassos

----------
From: Randy Read

Yes, the current version of Phaser will do the same test that xtriage carries out, and if it finds a sufficiently high non-origin Patterson peak, it will automatically characterise the translational NCS and use this for molecular replacement.  This is working pretty well in our tests.

In the near future you will be able to get the current version of Phaser as part of the upcoming CCP4 release, but at the moment the easiest way to get it is to download a recent version of Phenix.  You should be able to run that through ccp4i by downloading and installing the updated GUI files from our website (and getting ccp4i to interpret the command "phaser" as "phenix.phaser").

Best wishes,

Randy Read

----------
From: Randy Read

Airlie points out that what I said about the ccp4i interface wasn't correct!  In order to keep the ccp4i interface in synch with the version of Phaser, we've started distributing the ccp4i files with the source code.  The ones on our website are for an older version of Phaser, but the latest ones will come with the Phenix download that gives you the latest executable.

Apologies to anyone who was quick enough to download and install the wrong ccp4i files already!

Best wishes,

Randy Read

----------
From: Eleanor Dodson

That difference density looks too good to be true..

There is obviously a very strong non-crystallographic translation. Is the spacegroup and/or cell correct?
Eleanor

----------
From: xiaoyazi2008

Dear all,

Thank you very much for all the great suggestions on my case. 

Yes, I run the latest version of Phaser in Phenix. The analysis showed that there is one non-origin distinct peak more than 15 angstroms from the origin. 

44.1% origin:    FRAC 0.000 0.042 0.500   (ORTH  -15.7    2.8  103.5)

I managed to find four copies with the latest Phaser. After 50 cycles of rigid body refinement and 50 cycles of jelly body refinement, Rfree/R goes around 55/52. It is really hard for me to do model building at this point, because there is pretty much no new density. 

Compare to model building and refinement with normal dataset (no pseudo translation NCS), are there any special tricks or tips for structure determination from dataset with pseudo translation?

Thanks again!

Have a nice evening or morning or afternoon!

Zhihong

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

Dear Zhihong,

 

Refinement stuck at 55% Rfree (which is essentially random), means that you do not have found the correct MR solution. For me the prime suspect is the space group. In my experience pseudo translation or any almost crystallographic NCS will easily confuse automatic space group determination programs like pointless and it is often trickey to find out which symmetry is crystallographic and which is non-crystallographic.  

 

Since P21 is fairly low symmetry, I would reprocess your data in P1 and just naively run all your favorite molecular replacement programs (phaser, molrep, epmr, phenix, etc.) in P1. I would try them all since there are subtle difference between these programs and one may succeed where the others fail. Once you have a solution which refines (Free Rfactor drops below say 40% in P1) you can try to figure out what the true, higher symmetry space group may be. Your true space group may even be P1, with pseudo P21 symmetry!

 

Good luck!

Herman 

----------
From: xiaoyazi2008

Thanks all!

In addition to the pseudo translation problem, the model I used for MR is really a partial model with low quality. It could make it more complicated. I guess the MR solution is right since phaser gave very high Z-score (up to 60, maybe it is not real, too high to be true).  

It is possible that P1 is the hope. 

Are there any cases that structures were determined only by experimental phasing with pseudo translation? 

Nice weekend!

zhihong

----------
From: Randy J. Read

If you've used the older version of Phaser distributed with CCP4 then the TFZ for the second copy is not meaningful in the presence of tNCS.  Just putting something in the same orientation with the right relative translation will explain the modulation of the data so it will score well. However if you get a large TFZ with the latest version of Phaser then this means something, because it's already accounting for the modulation. 

Best wishes,

Randy Read

----

Randy J. Read

----------
From: Eleanor Dodson


pseudo translation is very common and usually does not pose a serious problem. you have to be careful about assigning the space group, since a translation od x,y,z=1/2 will generate absences for l = 2n+1 which may suggest a screw axis along c . But since your space group is P21 that isn't a problem here.

Your next problem is how to rebuild the structure!

Eleanor

----------
From: Ed Pozharski


One comment I'd like to add here is that in the presence of
pseudo-translational ncs that is nearly colinear with crystal axes you
will have a significantly higher R-value.  This may be a serious problem
with some reviewers when your R~30% on a 2A dataset.  It is completely
justified then to have the R-values calculated excluding absences.
Perhaps refinement programs should do this automatically if the
pseudo-translation is detectable? (perhaps as advisory output)