
Hi Patrick Take a look at the Binet-Cauchy kernel, e.g. here: http://bioinformatics.oxfordjournals.org/content/30/6/784.long It is a shape-correlation metric that is rather insensitive to global shape changes, but sensitive to local ones. Maybe that's what you need. It's also much faster to compute than RMSD. I have attached a python code snippet that computes a BC distance matrix for a set of proteins defined by CA coordinates. Kind regards, Dmytro. On 02/07/14 17:15, Patrick. C wrote:

Hi Patrick, RMSD is a poor measure in this case as it does not account for B-factors, occupancies, alternative conformations and so on information a crystal structure model may make available. Macromolecules are not a bunch of points in space. While I'm sure more thorough methods exist, I would vote for the simplest, most direct and obvious one. You can calculate electron density map using a Gaussian approximation from model A and B (yes, electron density map - not a Fourier image of it!). That will naturally account for all: B-factors, occupancies, other disorder. Then you can calculate a map similarity measure, such as map correlation, for instance. After all, why use a cannon to kill a fly?! If you are interested to follow this route I can explain the details. All the best, Pavel

I agree that would be useful as an alternative to RMSD, but if I understand the original post, the problem with RMSD is that the two secondary structure elements are connected by a variable turn so that they cannot be superimposed simultaneously. That could still be a problem, comparing maps. What you can do is report the change in angle between them, and the residues making up the hinge. A program called dyndom (dynamic domains) is good for this, Or you can superimpose each "domain" separately, view the superimposed molecules, and see haw far into the turn from each side thesuperposition is good To get the change in angle between the two parts, first superimpose model A on model B using only residues in domain 1 (say, the helix). Save that reoriented model A, and now superimpose it on model B using only residues in domain 2 (the strand). The angle involved in this second rotation is the chang in interdomain angle. (You could also report RMSD for superposition of the individual domain, but helix-on-helix or strand-on-strand are likely to be pretty good fits and not very informative.) eab On 07/05/2014 10:22 AM, PC wrote:

Hi Ed, yes, this makes sense, sure: having models to be superposed appropriately to begin with certainly is a good idea. My point was that coordinate-based rmsd does not account for B-factors and occupancy that is a problem if you are comparing flexible molecules. A possible solution may be to use a more generous (=more information rich) representation of atomic model such as electron density map that would 'automatically' take care of disorder. Pavel On 7/5/14, 9:37 AM, Edward A. Berry wrote:

Hi Patrick, supposing models are superposed appropriately (see previous comments on phenixbb), step-by-step I see it as following: - compute density map for model A (or its Fourier synthesis of resolution 1A or higher), - compute density map for model B (or its Fourier synthesis of resolution 1A or higher), - compute map CC per residue or per atom between maps corresponding to A and B. I would think that "superposed appropriately" is a key here. Pavel On 7/5/14, 7:22 AM, PC wrote:

Hi Patrick, why don't you superimpose only the matching segments and report their RMSD? It is the common procedure for RMSD's from superpositions to report the aligned residues together with the RMSD. The advantage compared to a map CC is similar to that of R_sym over R_meas: readers have a better concept (from experience) of what the numbers mean. Best, Tim On 07/02/2014 05:15 PM, Patrick. C wrote:
-- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A

Hi Tim, consider two atoms located at say 1A distance apart. In one case their B-factors are 1A**2 and in the second case their B-factors are 100A**2. In the first case these are two individual atoms and measuring distance between them makes sense, while in the second case atoms virtually coincide (within the cloud of their possible locations, given their B-factor) and therefore measuring distance between them isn't very meaningful. Using the map would differentiate these two scenarios, while usual way of computing RMSD would not. I guess this is what I was trying to say in my previous post. Whether the values of a metric of choice are intuitive or not is a separate question. After all you can always calibrate your expectations using examples with known answer. Pavel On 7/7/14, 3:47 AM, Tim Gruene wrote:

Hi Pavel, if the B-value would correspond to a cloud of possible locations, one should be able to move an atom within the cloud without altering the R-value and other quality indicators and without a refinement program moving the atom back to where it came from. In my experience this is not the case, why I prefer the (original) scientific interpretation of the B-value as to model of harmonic motion rather than its statistical. That is also the reason why I am not a friend of exorbital B-values in PDB-files. I think they tend to fudge poor modelling. Cheers, Tim On 07/07/2014 10:02 PM, Pavel Afonine wrote:
-- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A

That's a reasonable approach and I think that it is similar to the one used by FATCAT - which I notice is the basis for structural comparison at the RCSB site. Motivation and some results in the paper ( http://www.ncbi.nlm.nih.gov/pubmed/14534198) It's not clear to me, though, how to fairly compare the resulting RMSD of fragments with twists between. If I introduce an arbitrary number of twists then I can improve the rmsd artificially. In sequence matching there is a penalty for introducing a gap and that is scaled compared with the amino acid substitution scoring to split the match into a 'reasonable' number of sub-alignments. Obviously in 3D case there should also be a penalty for introducing a split in the structure to do a twist re-orientation - but how to quantify it compared with RMSD and get a global score? Seems to me better would be to express the whole problem in torsional space - so the twists would be large displacements while the matched sections should have close fit in torsional angles. And a global score could be calculated. Someone must have tried this? All the best Martyn Cambridge ----Original message----
Hi Patrick, why don't you superimpose only the matching segments and report their RMSD? It is the common procedure for RMSD's from superpositions to report the aligned residues together with the RMSD. The advantage compared to a map CC is similar to that of R_sym over R_meas: readers have a better concept (from experience) of what the numbers mean. Best, Tim On 07/02/2014 05:15 PM, Patrick. C wrote:
-- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A _______________________________________________ phenixbb mailing list phenixbb@phenix-online.org http://phenix-online.org/mailman/listinfo/phenixbb

Hi Patrick,
The quick and easy way of doing this would be to use Rob Nicholls' ProSMART
in CCP4. It does a local structure alignment and gives you a value for the
superposition of main chain atoms.
I don't know of any tool in Phenix that does this.
Cheers,
Morten
On 8 July 2014 14:56, MARTYN SYMMONS
-- Morten K Grøftehauge, PhD Pohl Group Durham University
participants (9)
-
C
-
Dmytro Guzenko
-
Edward A. Berry
-
MARTYN SYMMONS
-
Morten Grøftehauge
-
Patrick. C
-
Pavel Afonine
-
PC
-
Tim Gruene