revisiting rebuild

Case number:699969-2000875
Opened by:Enzyme
Opened on:Saturday, June 27, 2015 - 13:38
Last modified:Sunday, June 28, 2015 - 07:23

Lately many people have heard me complaining about rebuild. The exact problem is difficult to identify so I'll lay out a few of the symptoms here.

1) rebuild frequently attempts to place hydrophiles in the core as it's first option even if hydrophobes are available

2) Many times I have seen one of the shadow rebuilds placed exactly where I wish, but the rebuild never plces itself there. My understanding is that rebuild attempts valid positions while discarding invalid ones. But all too often the placement it allows are just as invalid as the positions it skips resulting in the "broken" backbone.

3) Banding during rebuild. For me, banding during a rebuild provides only catastrophic results. If others have success with it I would love to hear their technique, but if no one can successfully use it, I recommend that the bands on a rebuild be used as more of a suggestion to the rebuilder than the absolute position it attempts. Many times it will not even attempt to turn the segment from the first position where lands

Again I cannot say exactly what is wrong with rebuild, but I feel that it could use a good review to see if improvements can be made.

(Sat, 06/27/2015 - 13:38  |  5 comments)

bertro's picture
User offline. Last seen 24 weeks 2 days ago. Offline
Joined: 05/02/2011
Groups: Beta Folders

I would like to learn about the criteria used by the rebuild function to determine a `good` versus a `bad/skip` position. Why the `bad` position is rejected or the `good` kept?

brow42's picture
User offline. Last seen 1 year 37 weeks ago. Offline
Joined: 09/19/2011
Groups: None

I agree, banding during rebuild gets me 100% rejects, I can only use it to keep rebuild from throwing a strand completely out of the protein, not to guide the background. I have a script that is supposed to rebuild until a cis bond appears, it never works. I can't use a banded rebuild to tame glycines. I utterly fail at Snow Flea .

Berto, not exactly your question but: .

bertro's picture
User offline. Last seen 24 weeks 2 days ago. Offline
Joined: 05/02/2011
Groups: Beta Folders

from (thanks brow42 and katfish)

Rebuild tool

Thanks everyone for a great chat! Finally, we wanted to post the developer answers to the questions about how rebuild works:

1. The program looks at the selected sequence, looking at which amino acids and secondary structures are present, and builds a fragment library based on this.
2. The program picks a random subsequence of 3 continuous segments in the selected region and inserts a cut at one end of the selected region.
3. The program then does a look up into the fragment library for protein fragments that have similar AA and SS to this 3 continuous segment sequence and selects one.
4. The program then copies this segment onto the current protein.
5. The program applies an algorithm called CCD to get the new endpoints of the cut close together, and then closes the cut.
6. If the protein is messed up badly (scores poorly) it resets back to the original shape.
7. The process begins again at step (2).

Regarding 6.: Maybe this "scores poorly" could be a player parameter for when we want more or less positions retained?

Susume's picture
User offline. Last seen 3 days 11 hours ago. Offline
Joined: 10/02/2011

Rebuild with bands worked very well prior to NewChapter - within a few rebuilds it would jump to a shape that satisfied the bands, or as near as backbone lengths would allow. Since NewChapter it has been useless. It both fails to satisfy the bands and produces only horrible poses. I just gave up using it and use cut and move instead. Until now I think no one has been motivated enough to make a feedback about it.

jeff101's picture
User offline. Last seen 7 hours 31 min ago. Offline
Joined: 04/20/2012
Groups: Go Science

I often use bands with recipes that use rebuild.
I think all of these recipes disable the bands
before and re-enable the bands after each rebuild.


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