Serious issues rebuilding helices

Case number:699969-993805
Topic:General
Opened by:auntdeen
Status:Open
Type:Bug
Opened on:Thursday, October 25, 2012 - 21:59
Last modified:Wednesday, October 31, 2012 - 04:14

In working on the GUI script only puzzle in devprev, I've encountered a problem that I'd seen before recently.

Hand rebuilding of helices (at least) has serious issues. No matter the CI used (I've used both .05 & 1.0) and whether or not BB constraints are checked, helices can be either extremely slow in finding a reasonable pose, or will not do it at all - in many, but not all cases.

I've noticed this (and have been hearings grumblings about in both team and global) since the CASP roll puzzles started. It may date to the IE filter, or be a by-product of the BB constraints exacerbated by the IE filter.

I went back to the Platypus puzzle, where I recall was the last time rebuild worked normally to form a helix. I reset the puzzle, and tested with checked/unchecked, .05CI and 1.0 CI. I used the helix on the free end - and added some of the loop to it. Within a normal rebuild timeframe, it worked in all cases (see pix below).

I then moved on to 646, used what loaded, and tried to make a helix out of one end. Worked well.

On to 643. From the denovo chain, made a helix on an end no problem. Then on that puzzle, did an alignment, clicked on a yellow cutpoint, and tried rebuild. Fail (look at the distortion in the center of the helix in the second pix of that puzzle).

Came back to the GUI Scripting only puzzle, and tried all CI with BB checked and unchecked. The helix at segs 63 - 76 simply will not rebuild properly. I've tried doing it under all circumstances - the entire helix, just that one nasty end - no matter what, it will not form properly at any pose.

In the pix below I have the Undo graph open - you can see how fast you can get a helix shaped where it works, and how long you can let it go on when it is not working with bad results.

I am hoping that others who have encountered this problem will post, so that it can be nailed down.

One other note to help diagnose: apparently, scripts are not encountering this problem. I know that seems bizarre, but it is what it is.

Platypus pix:

http://fold.it/portal/files/chatimg/irc_99169_1351179020.png
http://fold.it/portal/files/chatimg/irc_99169_1351179033.png
http://fold.it/portal/files/chatimg/irc_99169_1351179073.png

Puzzle 646:

http://fold.it/portal/files/chatimg/irc_99169_1351179351.png

Puzzle 643:

http://fold.it/portal/files/chatimg/irc_99169_1351179498.png
http://fold.it/portal/files/chatimg/irc_99169_1351179676.png

GUI No Script Puzzle:

http://fold.it/portal/files/chatimg/irc_99169_1351180095.png
http://fold.it/portal/files/chatimg/irc_99169_1351180201.png
http://fold.it/portal/files/chatimg/irc_99169_1351180836.png
http://fold.it/portal/files/chatimg/irc_99169_1351181801.png

(Thu, 10/25/2012 - 21:59  |  17 comments)


Joined: 05/28/2012

Interesting enough while trying to rebuild by hand in normal client, I've had issues on 645. i had a piece of loop as well, but only the loop part was being rebuilt. It stopped me completely from exploring one of my tracks.

jflat06's picture
User offline. Last seen 5 days 2 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

The residueIE filter only affects the code in the design puzzles where we've enabled it, so I do not think that is the culprit. There's the potential that it is related to the backbone constraint checkbox (but if this is the case, it should be a problem on all puzzles). There's also the possibility that is somehow inherent in the starting structure or sequences of these recent puzzles, as this has caused similar issues in the past.

beta_helix's picture
User offline. Last seen 3 hours 29 min ago. Offline
Joined: 05/09/2008
Groups: None

Does it occur on Puzzle 644: Photo-Electric Hydrogenase Catalyst 1?

Joined: 04/19/2009

Some of my team members have reported this, that a helix will bend in the middle. I can say for myself that no matter what I did, my attempts to incorporate the entire length of movable loop (presented as sheet) by turning it into helix, refused to utilize the last 2 segments next to the frozen seg - they remained misshapen.

Joined: 04/19/2009

Yes - it could be co-incidence, but how then does rebuild work well in a puzzle if used on the denovo chain - but not work on a free strip if that chain is aligned? See the pix for puzzle 643 above.

utaca's picture
User offline. Last seen 15 weeks 3 days ago. Offline
Joined: 01/18/2012
Groups: Void Crushers

I want to add a very crazy helix. I saved it on my computer but forgot to upload it. Now I can't upload it because the puzzle is closed. Here is a screenshot:

brow42's picture
User is online Online
Joined: 09/19/2011
Groups: None

I have to agree that long helices won't rebuild, even if cut points leave the ends free. Literally, some segments don't ever change, and segments that have already been rebuilt keep changing. I see this in CASP and other prediction puzzles. The only way I can get the entire helix to rebuild is to freeze the part that has been rebuilt already, and rebuild the shorter segment.

Sometimes I wonder if its related to how rebuild seems to hold one end fixed, and freezing a middle segment moves the boundary point.

Possibly unrelated, I'm attaching a screenshot of a helix with a band. The band is practically zero length, it should be trivial to displace the backbone atom a tiny bit, and then continue the rebuild. In this picture, a denovo sheet is being rebuilt into a helix, with a tiny band on the middle atom (only there to keep it from wandering from the symmetry point during rebuild). As you can see, half the segment is completely untouched. I don't remember which end was the N terminal. This is from way back in July.

Joined: 04/19/2009

Here is another screenshot from an earlier puzzle from a teammate:

http://fold.it/portal/files/chatimg/irc_29473_1350163790.png

spmm's picture
User offline. Last seen 32 weeks 3 days ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

yes I have noticed that you have to manually pull helixes into roughly the right shape or they will just blether about and be quite unable to rebuild themselves even on CI=1, even rebuilds of loops etc on CI=1 will just happily bury themselves into the body of the protein in much the same way you expect from a rebuild on CI+0

brow42's picture
User is online Online
Joined: 09/19/2011
Groups: None

I'm pretty sure rebuild ignores CI, at least since I joined a year ago. Would a dev please confirm this? Scripts like DRW change the CI all the way to zero, and if it has no effect, then people should know this, to save time when hand rebuilding.

Today I couldn't get a helix to rebuild. I had to rebuild it to a sheet, and fix all the former cutpoints (which are now pretty obvious) before getting even close to a good helix rebuild.

gitwut's picture
User offline. Last seen 33 weeks 7 hours ago. Offline
Joined: 05/18/2012
Groups: Contenders

It's odd that I haven't encountered a helix yet that I couldn't use rebuild for shaping, since I use it almost exclusively to form/re-form Helices. The times I've encountered problems, it was due to not having enough slack on the surrounding loops or when a helix and sheet were conjoined (no intervening loop segments).

Slack is usually easy enough to add, but when dealing with conjoined helix and sheet, I temporarily change one or two sheet segments to loops and then rebuild. Neither freezing the sheet nor cutting (without converting a sheet segment to loop) seems to work for me when they're conjoined. Have these methods failed on your puzzles as well?

Since I'm relatively new (5-2012), I've just assumed that this is the way rebuild and helices work.

Joined: 04/19/2009

@gitwit - you bring up an interesting point.

Yes, you must have enough slack to form a helix, and formation will fail if you don't. The series of pix above for the GUI No Script puzzle, however, shows a helix attempt on a free end - so there was plenty of room for the helix to form properly - usually those are the easiest to form.

Thanks for contributing to the conversation, every post here will help the devs nail down what is going on!

beta_helix's picture
User offline. Last seen 3 hours 29 min ago. Offline
Joined: 05/09/2008
Groups: None

Does anyone notice this issue when rebuilding a helix WITHOUT using cutpoints?

ie starting from an extended chain (as auntdeen showed in her initial post above: http://fold.it/portal/files/chatimg/irc_99169_1351179498.png ) does anyone ever have trouble with that?
or given a starting model but not using the scissor tool.

Thanks!

Joined: 04/19/2009

No cutpoints in the GUI no script puzzle (see pix above), and I didn't insert / use any. That's what finally made me realize that something was wrong.

brow42's picture
User is online Online
Joined: 09/19/2011
Groups: None

I tried to rebuild one of the helices in Chicken Villin that got badly kinked, but it would never fix the bent part. It was already in basically the correct position with the end loops aligned to guide. Had only used band and wiggle on the puzzle.

jflat06's picture
User offline. Last seen 5 days 2 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

Is this still happening in the most recent devprev?

Joined: 04/19/2009

I allowed the update and tested on the GUI No Script puzzle.

I reset the puzzle and started from the reset on the helix at the free end.

It was better. At .05 with or without BB constraints checked, I got a decent helix shape in the "normal" amount of time. At 1.0 it seemed to also work in the normal amount of time.

At .2 with BB constraints checked, it went a little haywire - took about 3-4 the number of poses that the others took to make a decent shape (with some very funky ones in between).

So yes, it is better - but may not be completely fixed. Hopefully some other players will test and post their results.

And btw - really appreciate being able to see the score on graph points! Not so fond of the new icons for Remove Bands & Disable.

Sitemap

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, RosettaCommons