Idealize does not set ideality to zero

Case number:845813-2000773
Topic:Game: Tools
Opened by:Susume
Status:Open
Type:Bug
Opened on:Friday, May 29, 2015 - 01:50
Last modified:Tuesday, October 4, 2016 - 12:40

I have 3 contiguous residues with ideality -8.3, -13.7, -6.3. I select them in selection interface and hit Idealize peptide bonds. That whole end of the protein moves, and then their idealities are -5.6, -10.4, -5.6. Why doesn't the idealize tool fix the bonds so the ideality is zero? How can I improve their ideality, if not with the tool that is made for that purpose?

(Fri, 05/29/2015 - 01:50  |  19 comments)


jeff101's picture
User offline. Last seen 18 hours 33 min ago. Offline
Joined: 04/20/2012
Groups: Go Science

In the Selection Interface, there are 2 buttons for Idealize:

Button 5, a paintbrush labeled "Idealize the secondary structure!",
appears when 1 or more segments are selected.

Button 2, a wrench & screwdriver labeled "Idealize peptide bonds!",
appears when 2 or more segments are selected.

What is each of these buttons supposed to do?
Do both buttons affect the ideality scores?
Do they work best for a certain number of selected segments?

Perhaps using Random Idealize 1.8 or 2.1.1 would give better results.
http://fold.it/portal/recipe/100243
http://fold.it/portal/recipe/100786

bkoep's picture
User offline. Last seen 2 hours 56 min ago. Offline
Joined: 11/15/2012
Groups: None

The "Idealize Secondary Structure" tool sets the torsion angles of the residue to that of a common helix or beta sheet.

The "Idealize Peptide Bonds" tool sets the bond lengths, bond angles, and bond torsions of the peptide bond between two residues. This is generally useful for correctly closing cut-points, which disrupts the peptide bond between residues.

The "Ideality" subscore of a residue reflects all bond lengths, angles, and torsions of a residue, and will is affected by each of these tools.

brow42's picture
User offline. Last seen 4 days 12 hours ago. Offline
Joined: 09/19/2011
Groups: None

I think "Idealize Peptide bonds" is the same thing as "Fix backbone issues" in regular interface https://fold.it/portal/node/995580 .

spvincent's picture
User offline. Last seen 20 hours 31 min ago. Offline
Joined: 12/07/2007
Groups: Contenders

My understanding is that the backbone and ideality terms depend on bond lengths, bond and torsion angles, and perhaps other things too. I doubt if it's always possible to optimize a sequence so that everything's ideal because of the constraints imposed by the fixed ends. A bit like trying to fit a carpet into an area that's slightly too small for it.

Susume's picture
User offline. Last seen 19 hours 31 min ago. Offline
Joined: 10/02/2011

If the end points were fixed that would make sense, but the idealize tool swings half the protein out of place, so it seems to be moving one of the end points - why not move it to a place that makes the bond lengths and angles work?

spmm's picture
User offline. Last seen 49 weeks 17 hours ago. Offline
Joined: 08/05/2010
Groups: Void Crushers

This has been the case for a while I shared a minus 14 idealise score on one seg with the scientists - amonth or so ago - even if I cut the section out and idealised it it didn't change, even tried on different wiggle factors. Rebuild on different CIs and frag gizmo did nothing either. it was in a fairly loose loop so nothing clever about the pose.

Imho segs get stuck in some cases - bad luck if they get stuck with a bad score.

Joined: 09/24/2012
Groups: Go Science

Same for remix. Moreover, with remix, there seems to be no link with the fragment score.
Remix does something but fragment score is going down (and the score too).
It's confusing.

bkoep's picture
User offline. Last seen 2 hours 56 min ago. Offline
Joined: 11/15/2012
Groups: None

I'm having trouble reproducing this behavior here.

Susume, in which puzzle did you encounter this? Could you share that solution (or another that behaves similarly) with scientists?

Susume's picture
User offline. Last seen 19 hours 31 min ago. Offline
Joined: 10/02/2011

On 1091, shared solution "1091 Idealize does not set ideality to zero" with scientists.

In selection interface, select segs 16-18. Click Idealize peptide bonds. Ideality on those segs changes but does not go to zero. Same thing happens with segs 35-37.

bkoep's picture
User offline. Last seen 2 hours 56 min ago. Offline
Joined: 11/15/2012
Groups: None

Sorry for the delayed response, everyone.

This issue has to do with proline residues, which are handled uniquely by Rosetta. Because proline has to maintained its closed ring, "idealizing" this residue does not result in bond lengths, angles, and torsions that are considered "ideal" by the Rosetta score function. As a result, players should not expect to see an Ideality score of 0 for proline residues.

Additionally, because a residue's torsion angles depend on atoms of both the previous residue and the next residue (four consecutive atoms define a torsion angle), this also affects the Ideality score of those residues on either side of the proline.

In puzzle 1091, there are prolines at positions 17 and 36, which is why you are observing this discrepancy for segments 16-18 and 35-37. I expect you will see similar results around other prolines in the puzzle.

I understand that this seems nonsensical (i.e., the Idealize tool yields an imperfect Ideality score), but this is expected behavior, and is unlikely to change.

Susume's picture
User offline. Last seen 19 hours 31 min ago. Offline
Joined: 10/02/2011
Status: Closed » Open

Seeing similar behavior on puzzle 1208 in new devprev. Many residues have slightly negative ideality (-1 to -2.5), and the idealize tool does not change them at all. There are no prolines in my solution. Are these small negative scores considered tolerable, and is idealize supposed to ignore them? The overall ideality score for the 75-residue protein, after idealizing the whole protein, is -75. This is pretty bad for a small protein with no prolines that has already been idealized (from experience I would expect this number to be in the -30 to -50 range).

Shared solution as "1208 idealize does not change ideality".

bkoep's picture
User offline. Last seen 2 hours 56 min ago. Offline
Joined: 11/15/2012
Groups: None

Thanks for sharing, Susume!

The Ideality score component has been up-weighted in recent design puzzles like 1208, so I'm not surprised this penalty is more severe than what you're used to. We've noticed some peculiar geometry in Foldit designs and are trying to figure out what the problem is. There could be an issue with the Idealize tool, and your shared solution should be helpful in tracking down the problem.

Susume's picture
User offline. Last seen 19 hours 31 min ago. Offline
Joined: 10/02/2011

Seeing this again (or still) in 1285, the No Rebuild puzzle. Shared solution "1285 Idealize does not set ideality to zero". This protein has already been idealized once and then wiggled on low wiggle power, and now has a total ideality of -88, which is pretty poor for a protein this size (even among recent design puzzles). If you select all and hit Idealize a second time, the total ideality goes to -44, not zero. I still think the Idealize tool should set ideality to zero.

There are no prolines. The protein has never had cutpoints, and never been rebuilt, tweaked, or local wiggled. I have used rama map drag, Ideal SS, shake, global wiggle (always on auto wiggle power, with and without bands, with and without portions frozen), remix (by hand and in script) and mutate (by hand and in script). Somehow that subset of tools was enough to introduce a total ideality score of -100 (after several scripts, before I idealized the first time), and now Idealize is not able to get it back to zero.

Susume's picture
User offline. Last seen 19 hours 31 min ago. Offline
Joined: 10/02/2011

I reset the puzzle (1285 No Rebuild), and the total ideality on the extended chain is -204. This is approximately -2.7 per segment, but all the segments have slightly different ideality scores. I was assuming the extended chain we were given had zero ideality, and that all the segments were the same, but not so.

I selected the whole protein, set all to sheet, and hit Ideal SS. Now total ideality is -205. The ideality scores are now uniform, but not zero.

I thought the reason for the No Rebuild puzzle was that we kept coming up with poor ideality. Starting us off with poor ideality seems counterproductive.

Susume's picture
User offline. Last seen 19 hours 31 min ago. Offline
Joined: 10/02/2011

Puzzle 1290 (No Rebuilding) starts off with terrible ideality too (-247). Idealize on the extended chain only brings it up to -245. Setting the whole thing to either sheet or helix and using Ideal SS (which I'm pretty sure used to fix ideality as well as creating the chosen structure) also only gets it to -245. In end game on one track I was able to get it up to -49 (not a stable pose though). If this subscore measures bond angles and bond lengths that are not supposed to change, why give us a starting position with lots of room for improvement in those angles and lengths?

Joined: 04/24/2014
Groups: None

I'm looking into this to see if I can rustle up a good explanation.

Joined: 04/24/2014
Groups: None

We're thinking of a number of solutions here to clear up some of the ambiguity, and possibly updating the tooltip. Nothing firm yet though.

bkoep's picture
User offline. Last seen 2 hours 56 min ago. Offline
Joined: 11/15/2012
Groups: None

Sorry again for the delayed reply. I've chased down a few experts, and have done a little digging myself, but still haven't quite gotten to the bottom of it...

It looks like the issue stems from improper hydrogen torsions. (In addition to bond lengths and angles, the ideality scoring also reports on bad dihedral angles.)

There are a few cases in which hydrogens along the protein backbone can move slightly out of place. It's not a huge problem, but there are tools in Rosetta to address it, and we should be able to incorporate those tools into Foldit sometime soon.

However, I'm also seeing that particular hydrogens on ILE and VAL sidechains are systematically misplaced. As above, the difference is slight enough that it shouldn't affect any model's validity from a scientific perspective. Nevertheless, there seem to be some small differences between the "ideal" torsion values and the coordinates used by Rosetta to create an ILE residue from scratch. I'm still unsure why there is any difference to begin with (and someone might have had a good reason for making it so), but this issue is very unlikely to be resolved anytime soon.

I'd like to reiterate that this is no cause for concern among players. The Idealize tool was never meant to completely level the ideality score. An ideality score in the range of 1-3 points per residue is entirely reasonable and lies well within the range of geometries observed in high-resolution crystal structures.

Susume's picture
User offline. Last seen 19 hours 31 min ago. Offline
Joined: 10/02/2011

Thank you for the detailed response!

It does seem related to certain sidechains - I had not noticed that before. The extended chain at the start of 1290 has ideality -247 (all isoleucine). When I change all to serine, ideality jumps to -9. All leucine: ideality -58. All valine: ideality -232.

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