ED alignment

Case number:699969-995632
Opened by:drumpeter18yrs9yrs
Opened on:Monday, July 29, 2013 - 11:39
Last modified:Wednesday, August 7, 2013 - 16:18

why is it that when you align to ED, the protein rotates 180˚?

(Mon, 07/29/2013 - 11:39  |  7 comments)

Joined: 09/21/2011
Groups: Void Crushers

Simple answer: That is because the alignment button is useless for alignment.
More complicated: it is a very hard problem to compute the best or even a near optimal position in the cloud. That would just take too long.

And that it rotates 180 is just a coincidence that might be caused by the fact that the helixes are a pretty well aligned in the same direction.

Joined: 02/08/2012
Groups: None

Wouldn't it just be best to not even have it?

Angus's picture
User offline. Last seen 2 days 3 hours ago. Offline
Joined: 06/04/2008
Groups: Beta Folders

It's only useful when you have totally lost your cloud, usually in the Z axis.

Joined: 09/24/2012
Groups: Go Science

May be suggesting something like "align ED to selected" if one selects an helix and a part of the cloud ? otherwise, it seems this button is useless.

Joined: 09/21/2011

bruno might be on to something. i wonder if that would work.

spmm's picture
User offline. Last seen 19 weeks 23 hours ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

In the olden days the align to density button would often give you a big jump in score - particularly if you were roughly close to the pose. This is still the case in the gallery ED puzzle currently open.

jflat06's picture
User offline. Last seen 34 min 59 sec ago. Offline
Joined: 09/29/2010
Groups: Window Group

The problem with ED alignment is that we have a protein with residues and an electron density cloud with voxel grid values. It isn't like aligning two poses where you have residue x in one pose matching residue y in the other. So we cant just try and line up corresponding residues.

What we do instead is derive point-clouds from the atom positions of the protein and the positions of the cloud voxels. Then we generate a co-variance matrix from each of these, and compute a transformation between the sets of eigenvectors of these two matrices.

That approach makes two assumptions in order to function properly -

1. The shape of the protein is already similar to the shape of the cloud. This is a big problem for two reasons. The first is that you have to already pretty much have the answer in order to align via this method, in which case it's not actually all that useful. The second, which is a big problem since we've started doing untrimmed densities, is that if there are bits of cloud that do not correspond to something in the current protein, the whole algorithm is meaningless.

2. The eigenvectors within each set have different magnitudes. This basically means that the protein can't be 'mostly' rotationally symmetric along any axis. This assumption sucks because of things like helical bundles, which ARE mostly rotationally symmetric. This is the one that explains the '180 degree flip' behavior that you are referring to, as Timo has said.

In short, the electron density Align feature sucks in practice. I'm actually renaming the button to "Center Protein on Density" as that seems to be the primary use of it at this point.


Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, Boehringer Ingelheim, RosettaCommons