Hangers in global wiggle

Case number:671071-2012104
Topic:Game: Other
Opened by:ichwilldiesennamen
Opened on:Friday, September 24, 2021 - 07:43
Last modified:Thursday, September 30, 2021 - 10:28

Especially in symmetry- and binder-puzzles I noticed that global wiggling seemed to "hang" quite often. After looking into it more, it became obvious that this had to do with the clash-slider-settings. I very frequently used global wiggle at all CIs set to 1 exactly. And for these settings hangers or rather freezing does occur. With this I mean that even though there are prominent clashes or unidealities in the protein, there will only be a score-increase in the first iteration or so and then the score will stall but the iters will continue counting up.
Further testing showed that this can be avoided when for example the backbone-importance slider is set not directly to 1.00 but 1.03 instead. With this I get for the same protein/setup raising score for each iteration as it should be. This workaround works well for me. But if you don't know this then you can get frustrated like I did. It is also noteworthy that with a second run of global wiggle with all CIs at 1, the score will increase again for 1 iter and then stall again. This whole behavior seems to occur reliably everytime after local wiggle was used before like in my worm-recipe.
This could also have to do with filters because I always have all filters on while wiggling and it runs also well when all filters are disabled. But of course then the glob. wiggle leads in most cases to bad results/BUNs etc.
Even though I have this workaround it would still be good that some programmer could look into this because this may be frustrating for anyone that does not know this like it was for me.

(Fri, 09/24/2021 - 07:43  |  2 comments)

bkoep's picture
User offline. Last seen 2 days 13 hours ago. Offline
Joined: 11/15/2012
Groups: Foldit Staff

Thanks for the feedback, ichwilldiesennamen! Can you share a solution that demonstrates this behavior?

I suspect this is part of a known issue that is related to Objectives. We have an idea about how to fix it, but it is a little tricky to implement, so that work is still in progress.

Joined: 03/28/2020
Groups: Go Science

Hello bkoep,
this problems seems to be generic. So you can use any of my shares on 2047 and do the following procedure:
-set all importance-sliders to 1.00 exactly and turn all filters on
-insert at least one intentional clash into the structure by for example mutating a sidechain to something larger.
-perform a few local wiggles. I use my worm-recipe for this typically. Preferably not exactly where the clash is (? I am not sure if this is a jard requirement).
-this somehow breaks something down internally. So a subsequent global wiggle will stall after about 1 iter even though there is still the clash in and there would be points to gain.

-if you would then set the bb-importance-slider to 1.03 or so then the global wiggle will run as it should until the clash is resolved.

I hope you can reproduce this. And thanks for looking into it, iwdn


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, Boehringer Ingelheim, RosettaCommons