[cctbxbb] Best way to compare two structures using cctbx?

Martin Uhrin martin.uhrin.10 at ucl.ac.uk
Mon Mar 4 06:52:30 PST 2013


Dear Jan,

this is an interesting problem and there are no perfect solutions.  I'm not
currently a cctbx user so I can't say whether such an algorithm is
implemented but I done some work on this myself and can maybe point you in
the direction of some approaches you could take.

* Sorted distances (possibly the simplest approach)

1) Calculate all r_ij distances from all the atoms in your unit cell to all
neighbours within a cutoff.
2) Sort the list of distances.
3) To get the 'difference' between the two structures calculate the rms of
the differences in distances between the two lists.

Generally this approach works well although it can give false positives (by
virtue of the fact that it's throwing away a lot of information)

* Histogram radial distribution comparison (real space)

See: http://www.ncbi.nlm.nih.gov/pubmed/20720316 for details.

Basically calculate the radial distribution histogram for each species pair
and then use some method to calculate the difference between the histograms
(e.g. rms differences, cosine distance, etc.)

* Frequency space comparison

See http://arxiv.org/abs/0805.1418 for details.

Similar to calculating the r-factor although in this paper they use a more
sophisticated approach that I don't fully understand.

* Finding ideal transformation between structures

Hundt's approach (CMPZ):
http://scripts.iucr.org/cgi-bin/paper?S0021889805032450

Lonie's approach (XtalComp):
http://www.sciencedirect.com/science/article/pii/S0010465511003699
for details and
http://xtalopt.openmolecules.net/xtalcomp/xtalcomp.html
to try the algorithm for yourself.

Basically the idea is to find the more ideal transformation to take you
from one crystal structure to another.  From the transformation you can
then pull out one number to get a difference or use it to map from one
system to another.

I think I'll stop there but I'd be curious to hear about what you find and
how you get on.

All the best,
-Martin


On 4 March 2013 12:06, Jan Marten Simons <marten at xtal.rwth-aachen.de> wrote:

> Hi,
>
> I think I've seen some code in cctbx which tries to put the similarity of
> two
> crystal structures into a number. Am I right on this?
>
> I want to have something similar:
>
> I'm interested in comparing the direct space atomic distances of two
> structures sharing the same unit cell parameters and space group and
> scatterers but different atomic coordinates for those. Now I'd like to
> have a
> mapping where I could compare the bond lengths between all scatterers which
> should eliminate any movements which might also be due to a shift in the
> origin of a P 1 cell. Also I'd like to have a value for mean, minimal and
> maximal displacements.
> The purpose of this is to check how good a structure approximation matches
> a
> reference structure.
>
> How would you suggest to take on this task? Or is there even some code
> related
> to this?
>
> Thanks in advance,
>
> Jan
> _______________________________________________
> cctbxbb mailing list
> cctbxbb at phenix-online.org
> http://phenix-online.org/mailman/listinfo/cctbxbb
>
>


-- 
Martin Uhrin                                                  Tel: +44 207
679 3466
Department of Physics & Astronomy               Fax:+44 207 679 0595
University College London                         martin.uhrin.10 at ucl.ac.uk
Gower St, London, WC1E 6BT, U.K.         http://www.cmmp.ucl.ac.uk
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/cctbxbb/attachments/20130304/19243dd7/attachment.htm>


More information about the cctbxbb mailing list