[phenixbb] [ccp4bb] Se SAD phasing map becomes worse after refinement

[Randy Read]

Dear Zhu,

Questions specific to Phenix should really go to the Phenix-BB, so I am cross-posting my reply there.  Here I’ll focus more on generic issues.  There are also CCP4 tools that you could consider and presumably other people will offer advice on those.

One point you raise comes up so much that we have an entry in our FAQ (https://www.phaser.cimr.cam.ac.uk/index.php/FAQ) about it: “Can I use Phaser to check the correctness of a model I have already built and refined?”  The answer is no, because once you’ve refined the model it becomes better than random at predicting the data and therefore achieves a high LLG score, regardless of whether or not it is correct.

In this case, you might be able to use Phaser to help complete the model if one of the two copies of the complex is more completely modelled than the other.  If, say, you had a model for one copy you could fix that and search for a second copy: this should work because the refinement didn’t know anything about the second copy.

Phenix.autosol attempts to determine the NCS operators, so you need to check whether that has succeeded.  If not, you might need to try some more manual approaches to defining the NCS, which would help a great deal in map improvement.  Tom Terwilliger might answer this in more detail (perhaps on the Phenix-BB), but I don’t think you can build protein and nucleic acid in the same job, so you should look at the documentation to see how to do that.

Good luck!

Randy Read
-----
Randy J. Read
Department of Haematology, University of Cambridge
Cambridge Institute for Medical Research     Tel: +44 1223 336500
The Keith Peters Building                               Fax: +44 1223 336827
Hills Road                                                       E-mail: rjr27 at cam.ac.uk
Cambridge CB2 0XY, U.K.                              www-structmed.cimr.cam.ac.uk

> On 24 Mar 2020, at 04:31, Zhu Qiao <jasonqiao03 at GMAIL.COM> wrote:
> 
> Dear All
> 
> I am sorry for the long context. 
> 
> I have one protein (252 AAs, 2 Met) bound to double-stranded DNA (24 bp) crystalized.  I collected the Se-Met data of the crystal in C222 up to 2.8 angstrom. the space group is confirmed by running the pointless. 
> 
> I used the Phenix.Autosol to find the heavy atoms and get a quite nice map after the density modification.  It seems there are two proteins and two DNA duplex are in one ASU. Phenix.Autobuild can only build less than half of the protein sequence into the map and fill in the potential DNA map with amino acids. The Rwork/Rfree is 0.40 and 0.46, with the map CC=0.60. If I do the MR with the initial model built by Autobuild, the result TFZ=40, LLG=200+, which suggests the partial correction of the initial model. 
> 
> Here is the problem. From the map, I can see one of my protein domain and a clear feature of DNA double helix. But whatever I go further for manual build using coot, like building the DNA double-strand into the map and building the resolved domain, the refinement statistics go bad with R free ~0.50. 
> 
> I am wondering what's going wrong and how come the refinement can't improve the R factor. 
> 
> I have attached the relevant photos. 
> https://drive.google.com/drive/folders/1dJ4kn7CEHkCL3sMcCJtBqa5OsVFQngBx?usp=sharing
> 
> 
> Sincerely
> Zhu 
> 
> ########################################################################
> 
> To unsubscribe from the CCP4BB list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1