Centroid devprev test puzzle bugs

Case number:845813-994916
Topic:Game: Tools
Opened by:Bletchley Park
Status:Open
Type:Bug
Opened on:Saturday, April 6, 2013 - 19:32
Last modified:Monday, April 22, 2013 - 20:13

- It is said that sidechains are represented by a single large atom, they are not.
- Sidechains should have only one position and shake should have no effect, sidechains are normal and can move around as they always do, shake massively improves score.

This thus defeats the purpose of the puzzle.

(Sat, 04/06/2013 - 19:32  |  26 comments)


Joined: 05/19/2009
Groups: Contenders

Centroid *mode* ? Where is that *mode* activated ?

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

Sorry, the centroid mode wasn't actually enabled on that devprev puzzle. I believe jflat06 puzled the puzzle out from devprev for now.

We'll post a centroid puzzle once the new devprev client goes out.

Centroid mode is activated per-puzzle. Instead of calculating all of the atoms of the sidechains, it estimates them as single values. You will notice sidechains are all rendered as spheres of varying sizes, so it should be pretty apparent when it's actually enabled.

marie_s's picture
User offline. Last seen 1 year 41 weeks ago. Offline
Joined: 05/18/2008
Groups: None

On the graphical proposition : I don't know how you can fold without seeing the AA.
The classical view have to be accessible even with no possibilities to shake the sidechains.

on wiggle sidechains : this tool is modified or not?

on wiggle : wiggle froze almost every timeeven with ci = 0
http://fold.it/portal/files/chatimg/irc_33153_1365838903.png
wiggle with bands move very quickly

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

Thanks for testing the puzzle out.

The atoms of the AA are only roughly approximated in centroid mode which speeds up the scoring of the protein. Because of this, there is no absolute position information for all of the atoms of the amino acids. Drawing them in any particular way would be misrepresentative. Because of this, they are just drawn as spheres, as that's basically how they are being approximated in centroid mode.

I will check with some people, I believe wiggle sidechains doesn't do anything in centroid mode. I'll get back to you on this though.

For wiggle freezing up, could you share a solution with yourself with the protein in that configuration and tell us the name? It's much easier for us to look into it with the exact pose.

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

These buttons should be disabled in centroid mode: Shake, Mutate, Wiggle Backbone, and Wiggle Sidechains.

They weren't all disabled in the selection interface. We've fixed that for the next version so it isn't confusing.

Susume's picture
User offline. Last seen 2 days 10 hours ago. Offline
Joined: 10/02/2011

If you click "Show expected residue burials" in the centroid puzzle, the client crashes. Latest devprev, Win 7 64.

end of Log.txt:

ERROR: AtomTypeSet: unrecognized atom parameter: SASA_RADIUS
ERROR:: Exit from: E:\vmuser\foldit\mini-interactive-main-release\rosetta_source\src\core/chemical/AtomTypeSet.hh line: 142
ERROR: Rosetta exit unexpectedly
SRVR_THRD getting notifications...

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

Thanks for catching that. I think 'show expected residue burials' is expecting some information that is not available in centroid mode. I'll fix that.

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

I've fixed this bug, it should go out in the next version. Assigning this feedback to Seth so he can look at the other issues.

marie_s's picture
User offline. Last seen 1 year 41 weeks ago. Offline
Joined: 05/18/2008
Groups: None

wiggle problem : testwigglecentroid

Information on segments are on tab, I want to see this information to fold the protein because I choose the shape of the protein with this information: not all segments are in corner, not all hydrophobes are outside...

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

Thanks for pointing this out. The reason it wasn't there was because the score terms used for centroid mode are different than normal so they were just being lumped into "Other". I've updated the code to add centroid terms to the appropriate categories. Just a note, you won't see all the same categories that you're used to with normal puzzles because not all of that data is being calculated so it's not there to show.

Joined: 09/14/2008
Groups: Void Crushers

I would love a 'centroid' tag so our script can 'skip' all the unnessecary shake.

Joined: 05/19/2009
Groups: Contenders

New centroid puzzle crashes previous client devprev upon wiggling.

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

Previous versions of the client will not be able to load in centroid puzzles. So that's to be expected. When this actually goes out to main the server will prevent previous client from trying to load in the puzzle, but it doesn't do that for devprev puzzles since they're just there for testing.

brow42's picture
User offline. Last seen 4 hours 43 sec ago. Offline
Joined: 09/19/2011
Groups: None

Puzzle comments (just from poking around, haven't played it yet):

All segments have 1 rotamer, except histidine, which has 2. This seems strange.

All segments have 7 atoms (except proline, which is missing the usual hydrogen). 5 and 6 are both sidechain atoms. Both are bandable by scripts, and presumably, both are bandable by hand as well. Perhaps 5 is close to the backbone and 6 is far away (and big)? Banding by hand might be unreliable, maybe you sometimes band to 5 and sometimes to 6?

Scripting comments:

Devs should make shake and wiggle sidechains return immediately, so that scripts do not have to be modified to skip these time-consuming steps.

Devs should provide a function to indicate centroid mode so that scripts can be modified. Should be in behavior or puzzle depending on whether users will be able to turn this on and off within a puzzle or not.

Most of the score parts are gone. Scripts like DRW should be modified to check which score parts actually exist (disulfide being an exception). I know Timo checks for Other part, but now he has to check for all the others.

Script workaround:

This will check for centroid and turn off sidechain shake and wiggle for Lua2. Lua1 code would be similar.

function IsCentroid()
return tostring(current.GetSegmentEnergySubscore(1,'sidechain')) == '-0'
end
function DisableSidechains()
local function null(i) return end
local function ReplaceWiggle(Wiggle)
return function(i,a,b) if a ~= false then Wiggle(i,true,false) end end
end
structure.ShakeSidechainsAll = null
structure.ShakeSidechainsSelected = null
structure.LocalWiggleAll = ReplaceWiggle(structure.LocalWiggleAll)
structure.LocalWiggleSelected = ReplaceWiggle(structure.LocalWiggleSelected)
structure.WiggleAll = ReplaceWiggle(structure.WiggleAll)
structure.WiggleSelected = ReplaceWiggle(structure.WiggleSelected)
end

if IsCentroid() then DisableSidechains() end

Joined: 05/19/2009
Groups: Contenders

Try running public script void crusher on this, you'll see the problem.
It does not move at all with wiggle...

brow42's picture
User offline. Last seen 4 hours 43 sec ago. Offline
Joined: 09/19/2011
Groups: None

I have seen this, but only with weak bands. Using Rav3n Compress 2.2.1, bands progress from 0.3 by 0.7...protein does not budge until .44.

Joined: 05/19/2009
Groups: Contenders

And what does it do as soon as the bands are removed ? I have not seen wiggle produce any movement without bands in this client on this type of puzzle.

brow42's picture
User offline. Last seen 4 hours 43 sec ago. Offline
Joined: 09/19/2011
Groups: None

I'm currently #2 rank on the puzzle. Wiggle seems to work fine.

Joined: 05/19/2009
Groups: Contenders

Not from my experience. My reporting of this bug stands.

Joined: 05/19/2009
Groups: Contenders

in client number of players will not reflect number of players on website, regardless of filter settings.

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

Is this something you have only seen with this centroid puzzle? Only on devprev puzzles?

Joined: 05/19/2009
Groups: Contenders

I have noticed it only on centroid puzzles. Will check again on others and post if I find anything on other puzzles related to wiggle.

Joined: 05/19/2009
Groups: Contenders
Status: Open » Open

Try this:

No bands.
Wiggle with cl = 0.00

NOTHING happens. This is in the latest client devprev 2013-04-18-54935M-win_x86-devprev.

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

In the initial pose the van der waals forces are basically 0 because the residues are already far enough away from each other. This is why moving the clashing importance down doesn't really make a difference when you wiggle straight off the bat.

Just a note: The tools and scoring of centroid puzzles is similar, but won't behave exactly the same as in normal puzzles. Some strategies will not work and some may need to be modified.

Joined: 05/19/2009
Groups: Contenders

client doing nothing uses 9% or more CPU.

tamirh's picture
User offline. Last seen 6 years 10 weeks ago. Offline
Joined: 05/11/2012

Even when the client is not actively performing any operations it's still rendering, so some CPU usage is to be expected.

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