Impossible H-bonds can be made on 1992.

Started by BootsMcGraw

BootsMcGraw Lv 1

When I ran a script on puzzle 1992 to make hydrogen bonds, I saw strange bonds forming between residues on opposite sides of the tetramer, clearly physically impossible. See link for a photo.

I didn't write down between which residues this occurred, but next time I see this, I will save the solution and submit for the scientists to review.

If anyone else is having this problem, please stop your script at that point, and save and share your solution. I see that frood66 had this same issue, so it isn't a glitch on my own machine.

ichwilldiesennamen Lv 1

Yes, I see these too.
The bonds themselves are OK as well as the bonus-function. But the blue/red markings are wrong. Seems like foldit cannot cope this symmetry correctly. These wrong markings occur on almost every larger net. If it helps I can post multiple examples of it.

HuubR Lv 1

Just my two cents of what might be happening here.

In a "regular" Symmetry puzzle, any H-bond you have between chain 0 (the "O"riginal) and chain 1 (the 1st copy), will also be present between chains 1 and 2, and between chains 2 and 0 (in case of a Trimer), or between 2 and 3, as well as 3 and 0 (in case of a C4 Tetramer).
I would not be surprised if the visualisation algorithm only looks at the first one, and then assumes that it needs to show the same bond in 2 or 3 other positions.  If I would be a programmer (which I am not), I might have programmed it in that way, because that is the simplest algorithm, and in many cases a simple algorithm is to be preferred over a complicated one.

Now we have a D2 Tetramer (see also my comment to 1992), and the situation is quite different. You can form a bond between chains 0 and 1, but that will not be present between the corresponding atoms in chains 1 and 2.  There will be four instances of this bond, but they will be in different positions. Let's say there is an interface bond between atom A in chain 0 and atom B in chain 1. When I write that as 0A-1B, the same bond should also be visible as 1A-0B (i.e. a "mirror image" in the same interface), 2A-3B, and 3A-2B. But it looks like it is now displayed as 0A-1B (correct), 1A-2B (which could be quite far apart), 2A-3B (again, correct), and 3A-0B (far apart).

In short: my guess would be that the visualisation algorithm is not aware of the fact that this is a D2 Tetramer, and it shows the bonds as if it were a C4 Tetramer.

BootsMcGraw Lv 1

My thoughts, exactly. But they seem to be figured into the calculations of H-bond scoring, when I think they shouldn't.

If these aren't legitimate H-bonds, then it is not helpful to the scientists for us to incorporate them into our designs.

bkoep Staff Lv 1

Thanks for the feedback! The screenshots are very helpful.

From what I can tell, the issue seems to occur only with the H-Bond Network Objective visualization? (These are the bright blue and red bonds that appear when you click the Show checkbox in the Objectives panel.) Has anybody observed this issue for the regular blue/white striped H-bonds that are toggled in the View Options menu?

From the couple of solutions that I've checked, everything seems to be scoring correctly. I believe this bug is strictly a visualization issue.

BootsMcGraw Lv 1

I have not seen these with normal H-bonds, only between monomers.

So the crazy bond is visualized between the wrong sidechains, but it is correct? [head explodes]

HuubR Lv 1

Yes, bkoep, you are right that the issue shows only with the blue and red visualisation of the HBN objective (i.e., with "Show" checked in the Objectives panel). The blue/white bonds in and around chain 0 are OK.
But I think your impression that the score is OK might not be correct in all cases.

First of all, I have to admit that my own assumption in comment #2 (above) was wrong.  When I wrote that, I had only seen screenshots; I should have waited until I encountered the phenomenon in my own design.

In this example, I have two Hydrogen Bond Networks, each consisting of (when I count them) 8 bonds, 6 of which are interface bonds. On one side, that is shown and counted correctly. But on the other side, it is shown and counted with 7 interface bonds.

When I mess up the HBN, it falls apart into two separate bonds, so it does not count as a HBN anymore. But on the "Fake" side it is still shown, in blue, with 3 interface bonds. The score is zero now, because this HBN is (much) less than 75% satisfied.

When I tried to make a bond to the Acceptor side of the Threonine, the other bonds got messed up again, but to my surprise I saw a pair of fake blue bonds on the other side (the side that had previously been OK), as well as the "original" pair of fake bonds, but now red because they had become too weak. Note: I rotated the camera to the other side for this screenshot.

Instead of making another assumption, let me call it a hypothesis this time:
Is it possible that the fake bonds occur whenever a bond is formed between an Acceptor on chain 0 (coloured) and a Donor on chain 1 (purple)?  I think that description would fit all the screenshots I have seen so far, and any other combination seems to be displayed OK.

I have uploaded three solutions (with scores 22957, 21081, and 20913, as shown in these screenshots) for scientists. Hope it helps you track down this bug.

jflat06 Staff Lv 1

Hey all, thanks to your efforts, we've discovered a bug in Rosetta!

We're talking with some Rosetta devs and coordinating a fix that should hopefully come in the next few patches.

jflat06 Staff Lv 1

Just checking back in to say that we think we have a fix, and it should be coming in the next release :)

Thanks again for finding this bug!