Pi helix trick revisited

Case number:699969-991059
Topic:General
Opened by:infjamc
Status:Closed
Type:Bug
Opened on:Thursday, November 10, 2011 - 03:23
Last modified:Thursday, March 8, 2012 - 18:48

In the recently posted Beginner Symmetry puzzle, it's possible to turn the alpha helix into a pi helix for a huge boost in the score:

http://fold.it/portal/files/chatimg/irc_92665_1320894770.png

I don't think this was intended. For future purposes, could constraints be placed (say, require a minimum distance between the first and the last residues) on this type of puzzle to discourage the use of high-scoring but unnatural artifacts?

(Thu, 11/10/2011 - 03:23  |  20 comments)


beta_helix's picture
User offline. Last seen 5 days 21 hours ago. Offline
Joined: 05/09/2008
Groups: None
Status: Open » Open

Thanks for catching this infjamc, this is why we wanted to start with a simple first symmetry puzzle, to try to catch these bugs that went unnoticed in the Developer Preview.

infjamc's picture
User offline. Last seen 2 years 44 weeks ago. Offline
Joined: 02/20/2009
Groups: Contenders

Well, maybe bug isn't the right term-- I see it more as exploiting a known weakness in the score function, which gives more points to "stronger" hydrogen bonds (i.e. you get more points because the pi helix is more densely packed). In the interest of full disclosure, I'd say that this bug gives one at least a 400-point advantage (my score was in the 9600s before posting the screenshot).

Joined: 12/27/2010
Groups: None

Please enlighten me as to how pi helices are created. Presumably it is not simply a rebuild of a helix.

infjamc's picture
User offline. Last seen 2 years 44 weeks ago. Offline
Joined: 02/20/2009
Groups: Contenders

1. Enter a full-atom viewing mode that displays polar hydrogens (e.g. via the Ctrl+Shift+V hotkey on Windows systems).
2. Add a rubber band between the backbone hydrogen atom of residue n to the oxygen atom of residue n+5, starting with n=1. Then, set the length of the band to zero.
3. Repeat Step #2 for the length of the helix.
4. Wiggle, then shake. If done correctly, the alpha helix would curl up into a pi helix. (Note that this trick obviously doesn't work if there's a proline in the middle.)

Joined: 06/17/2010

Puzzle should be canceled?

infjamc's picture
User offline. Last seen 2 years 44 weeks ago. Offline
Joined: 02/20/2009
Groups: Contenders

Either that, or decrease the number of points it's worth.

On a side note, I've just discovered another bug. This has been confirmed by several of my teammates:

1. Load Puzzle 479.
2. Load a teammate's solution as a guide.
3. Press the G key. You should see the guide fly away, and you can't get it back unless you restart the puzzle.

beta_helix's picture
User offline. Last seen 5 days 21 hours ago. Offline
Joined: 05/09/2008
Groups: None

thanks infjamc, steveB mentioned this in the news post:
http://fold.it/portal/node/991056#comment-14004

Joined: 04/19/2009
Status: Open » Open

Changing status back to Open and priority to High:

I reset the puzzle. Started one script - killed it after literally 2 minutes. Started another, and within one minute my score jumped to rank #11 - with this unlikely result:

http://fold.it/portal/files/chatimg/irc_99169_1320986993.png

No way should this score so high. A teammate was able to duplicate this with the same results.

Sigh - puzzle should be cancelled.

We had both done the dev update this evening, if that has anything to do with it.

spmm's picture
User offline. Last seen 40 weeks 2 days ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

2. Add a rubber band between the backbone hydrogen atom of residue n to the oxygen atom of residue n+5, starting with n=1. Then, set the length of the band to zero.
3. Repeat Step #2 for the length of the helix.

I'm obviously not getting this, hydrogen = white; oxygen = red - starting from the glutamine at seg #1=n - which has 3 white knobs close together > to the Leucine at seg#6, which only has backbone bonds. alanine at 11, leucine at 16, glutamate at 21 etc.

The residues themselves don't have apparent red knobs to bond to, even with bondable atoms turned on.

with band length set to zero - then wiggle - ends up in a ball?

infjamc's picture
User offline. Last seen 2 years 44 weeks ago. Offline
Joined: 02/20/2009
Groups: Contenders

Oops, looks like I wasn't being clear enough. What I meant in Step 2 was this:

Starting from the alpha helix configuration, band the backbone hydrogen of every residue to the oxygen of the residue that's 5 amino acids ahead. (In other words: ignore the side chains completely.) Namely, 1 to 6, 2 to 7, 3 to 8, etc. Clashing importance and band strength should both be left at 1.00.

Maybe a screenshot from the starting configuration would help? (See attachment)

Joined: 12/27/2010
Groups: None

Thanks infjamc for your tips. It helps to turn sidechain view off as you have done in your picture otherwise it gets too confusing.

Joined: 12/27/2010
Groups: None

This little snippet of code seems to produce the bands for a pi helix - at least on this particular protein.

for x = 1, structure.GetCount()-5 do
band.AddBetweenSegments(x,x+5,3,1)
end

Is that correct?

beta_helix's picture
User offline. Last seen 5 days 21 hours ago. Offline
Joined: 05/09/2008
Groups: None
Status: Open » Closed

We hopefully fixed the pi helix bug, please let us know if it still scores better than regular alpha helices.

infjamc's picture
User offline. Last seen 2 years 44 weeks ago. Offline
Joined: 02/20/2009
Groups: Contenders
Status: Closed » Open

To whom it may concern:

It seems to me that the pi helix is still artificially scoring higher than the standard alpha helix in symmetry puzzles:

http://fold.it/portal/files/chatimg/irc_308854_1331164588.png

infjamc's picture
User offline. Last seen 2 years 44 weeks ago. Offline
Joined: 02/20/2009
Groups: Contenders
Joined: 09/21/2011
Groups: Void Crushers

Checked the segment scores of the helix in a very early stage, no straight bonds and not connected to the other one.
Total was then about 1060
In the shown picture it is about 1230

Not a big difference and can be explained by the extra bonding to the other one.

See the pictures with the segment scores of the first 10 segments.

If there is a bug it is not with the segment scores.

Joined: 09/21/2011
Groups: Void Crushers

I just did an experiment on an early version.

I did my best to create one, succeeded and the score did went up a lot.
Made sure there was no interaction.

So I must conclude the bug is there. Of course I did not know about this before, thought it was a bonus for the better bonding with the other one.

Joined: 09/21/2011
Groups: Void Crushers

A last remark:
The segment scores did not change by much about 50 because of other improvements.
Also the helix scored normal.

So the bug must be in the computation of the interaction bonus points.

infjamc's picture
User offline. Last seen 2 years 44 weeks ago. Offline
Joined: 02/20/2009
Groups: Contenders

Timo:

Thanks for taking the time to do the testing on your own. Your results are pretty much consistent with what my teammates and I have found-- namely, pi helices score too high because their backbone hydrogen bonds are being given too much credit. If I remember correctly from the developers, this is because the backbone hydrogen bonds in the "n to n+5" case is being scored under the wrong category of hydrogen bonds.

beta_helix's picture
User offline. Last seen 5 days 21 hours ago. Offline
Joined: 05/09/2008
Groups: None

Thank you for bringing this up.
We have fixed this bug (once and for all) and this will go out in a future release.

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