Core Exists

Case number:699969-997222
Topic:General
Opened by:wisky
Status:Open
Type:Bug
Opened on:Monday, March 10, 2014 - 05:17
Last modified:Thursday, March 13, 2014 - 20:53

There is a problem with the Core Exists filter. I have shared this solution (7981, "Core Exists!") to scientists. I have frozen FOURTEEN! (14!) core residues that are NOT being registered by the Core Exists filter. This is a HUGE problem, and it has been going on since the start of Core Exists. Four of the residues are arguably (in my mind) not core residues, but you can clearly see that at least 10 of them are core residues, but not being registered.

I think Core Exists is a good condition to have, but the way it's implemented is INCREDIBLY restrictive to many players. It is messing with recipes and hand folders, and will leave you with helix only solutions as your top scorers. Many players do not put any effort into design anymore for this very reason.

Fix this PLEASE. This has been going on since Core Exists came out.

(Mon, 03/10/2014 - 05:17  |  8 comments)


wisky's picture
User offline. Last seen 5 weeks 1 day ago. Offline
Joined: 07/13/2011

http://fold.it/portal/files/chatimg/irc_259088_1394428657.png
http://fold.it/portal/files/chatimg/irc_259088_1394428662.png

Two pictures for those who are curious. The frozen residues are being tagged by core exists as not being core residues.

Joined: 09/21/2011

It also doesn't wiggle. Any time where wiggle all comes up in script, the protein just sits there locked up. Blue fuse can drop the score mildly and it bring the score back up. I think this also has to do with the core filter being way too picky (just a gut conclusion).

wisky's picture
User offline. Last seen 5 weeks 1 day ago. Offline
Joined: 07/13/2011

This is what I was getting at, albeit indirectly.

bkoep's picture
User offline. Last seen 9 hours 3 min ago. Offline
Joined: 11/15/2012
Groups: None

The filter shouldn't affect wiggle. In fact, the wiggle tool doesn't even know about the filter, which is why wiggle can inadvertently lose points from the filters. We've been getting reports that wiggle has been a little sticky lately, so this is probably a more general issue.

Joined: 09/21/2011

Does the wiggle also not know about the score? I ask because even the slightest movement of the protein was causing my 3 helix model to plummet in score due to core penalties. My turkey replica version would plummet at the slightest movement because of fragment filter penalties.
This is also the first and only time I set up any design protein and ran drw all night to only receive 36 points. When I tried working the protein by hand, I would hit wiggle all and nothing would happen. When I watched scripts run and they reached the wiggle all portion, nothing would happen. This puzzle had the most pronounced sticky wiggle I have ever seen. It may be a coincidence that the filters are also peculiarly touchy.
If this is the case then why is the design puzzle affected so deeply by the wiggle issue? Maybe because mutate side chains locks up the wiggle?

bkoep's picture
User offline. Last seen 9 hours 3 min ago. Offline
Joined: 11/15/2012
Groups: None

Wiggle only knows about the "score function" part of the score—that is, the score that is calculated before applying filters. We're looking into some possibilities about why wiggle seems to be sticky, but there is no glaring reason why this puzzle would be worse than others.

I know it's more than you asked for, but here is the reason that wiggle doesn't know about the "whole" score:
The basic score function is itself smooth and continuous, which means that we can use mathematics to find the score function's gradient (the direction in which the function goes up). This is how wiggle works; it calculates the gradient of the score function and moves the protein in that direction. Wiggle stops finding points when it reaches a local maximum in the score function (i.e. all directions lead downward).

Filters are different. They are not continuous, because they change from "pass" to "fail" in an instant. Unlike with the score function, wiggle cannot figure out mathematically whether the filter will pass or fail before it makes a move. We could make wiggle backtrack whenever it causes a filter to fail, but that would result in some REALLY weird (and inefficient) wiggle behavior. In fact, the ability to solve filters is precisely one of the advantages we think Foldit players have over computers.

bkoep's picture
User offline. Last seen 9 hours 3 min ago. Offline
Joined: 11/15/2012
Groups: None

Thanks for sharing this with us, wisky. I was just looking at your solution, and the Core Exists filter seems to be working correctly for me. The solution does have some core residues (it actually looks like 18 or 19 are registering as "core"—the way Foldit identifies "core" residues is pretty generous, as you noted). But this solution is still a few residues short of meeting the filter threshold: 30% of the total residue count need to be in the protein core.

At the lab, we've been very happy with the cores that players have designed with these filters. But if the filter seems especially restrictive to design, maybe we can try reducing the threshold of this filter or reducing the per-residue penalty for falling short of the threshold.

brow42's picture
User offline. Last seen 1 week 2 days ago. Offline
Joined: 09/19/2011
Groups: None

The shorter the design, the less restrictive the filter has to be. It's very hard to make a valid core with 50 residues. Expect at least a 300 point penalty, the comes and goes at the slightest touch.

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