[phenixbb] question on molecular replacement

Pavel Afonine pafonine at lbl.gov
Mon Jun 11 12:49:59 PDT 2012

Hi Fengyun,

it sounds like a perfect candidate for Arcimboldo. Last week at Erice 
school I've seen Isabel solving quite a few structures that people could 
not solve before using conventional methods.

The best way is to contact Isabel directly (as it is not too 
straightforward to use Arcimboldo in its current state) and it's likely 
she solves it.


On 6/10/12 9:39 PM, fn1 at rice.edu wrote:
> Hi Nathaniel,
> The crystal reflects to about 2 A. Phenix.xtriage indicates there's no 
> twinning. The protein is only 72 aa long by de novo design and we have 
> a simulated model. This protein should form a three-helix bundle like 
> structure. But i am not pretty sure how accuracy the model is.
> Right now i am using phenix.automr to do all the stuff. I don't have 
> rosetta installed yet.
> Thank you!
> Fengyun
> Quoting Nathaniel Echols <nechols at lbl.gov>:
>> On Sun, Jun 10, 2012 at 7:57 PM, <fn1 at rice.edu> wrote:
>>> I have a dataset in P21221. The cell content analysis shows that there
>>> should be 5 monomers in asu. The gel filtration profile of this protein
>>> shows that it should form a tetramer.
>> I agree with Francis - try all possible space groups in the P222 
>> point group.
>>> I have a simulated structure as searching model. I try to use the whole
>>> model or backbone model, and different resolution cutoff. When the 
>>> whole
>>> model is used, the llg of solution is negative. When the backbone 
>>> model is
>>> used, the llg is about 160 when searching for 4 monomers and tfz is 
>>> 7.4. But
>>> the following autobuild process could not lower rfactor than 50%.
>> It would help if we knew more about the properties of the model and
>> data.  What is the sequence identity of the closest homologous
>> structure?  Comparative models are often problematic, and those
>> statistics aren't great.  Also, if the resolution is poor, AutoBuild
>> isn't going to be able to do much.  This may be a case where
>> MR-Rosetta is the solution, but I'd verify that you're running the MR
>> search correctly (and that it has any chance of working to begin with)
>> before trying that.
>>> I notice online that it might be useful to run self rotation 
>>> function first
>>> to find ncs operators before doing mr. Is this true and how can i do 
>>> this?
>> I'm pretty sure CCP4 has a program to do this, but I don't think we
>> have anything in Phenix yet.  (I suspect it's straightforward to code,
>> but I couldn't figure out the math when I looked at it briefly last
>> year.)
>> -Nat

More information about the phenixbb mailing list