Tournament Phase Two puzzles reposted!

The Partition Puzzles released yesterday were posted with the wrong score function, unfortunately. The old puzzles have been pulled and reposted with the correct score function. Players may load their work from the previous puzzles into the reposted puzzles. We're sorry for any inconvenience (but we're glad we caught the error early)!

Also, we'd like to apologize to LociOiling, whose tournament submission was misattributed to Newton76. This has been corrected in the blog post, and the reposted Partition Puzzle here. Sorry LociOiling and Newton76!

(Tue, 09/18/2018 - 16:03  |  16 comments)
frood66's picture
User offline. Last seen 2 hours 45 min ago. Offline
Joined: 09/20/2011
Groups: Marvin's bunch
new scoring

Many longer serving players will remember how the exploration score affected scores in the past - or, at least, generally how this worked.

Perhaps it would be a good idea to explain how the exploration score affects overall score in these partition puzzles?

It could be helpful to all players (both old and new) I think.

Just a thought.

jausmh's picture
User offline. Last seen 6 hours 44 sec ago. Offline
Joined: 05/19/2017
Groups: Marvin's bunch
new scoring

I think so as well.

Maybe we could get to know how the exploration score was handeled before oposed to how it is now. I.e. how the distance is actually computed.

Susume's picture
User offline. Last seen 1 hour 34 min ago. Offline
Joined: 10/02/2011
Exploration is not a score

Looks like Exploration is not a score on these puzzles - you don't get more or less points based on meeting it. It's just a yes-no condition that redlines your current score if your protein backbone shape is not different enough from the original design. When the score is redlined, you don't get credit for it. As soon as you remove the red line (by getting RMSD above 2.5 A), you get credit for the score.

From Wikipedia: "In bioinformatics, the root-mean-square deviation of atomic positions (or simply root-mean-square deviation, RMSD) is the measure of the average distance between the atoms (usually the backbone atoms) of superimposed proteins."

frood66's picture
User offline. Last seen 2 hours 45 min ago. Offline
Joined: 09/20/2011
Groups: Marvin's bunch
I really look forward to a

I really look forward to a reply from FC - I appreciate yr input Sus - but I think many want FC to clear this up.

I could have written what U just have - It's just not the info (or the origination) that many would like.

Thx anyways Sis - I know where Yr coming from :)

bkoep's picture
User offline. Last seen 1 hour 52 min ago. Offline
Joined: 11/15/2012
Groups: None
Unlike old-school "Diversification Puzzles"

In the Partition Puzzles, the Exploration Objective simply uses RMSD to measure the "dissimilarity" between a pair of structures: your current structure and the starting structure. The idea is that we are only interested in solutions that are dissimilar from the starting structure. So if a solution is too similar to the starting structure (i.e. RMSD < 2.5 Å), the solution is invalid and its score will not be registered with the leaderboards. If you've played any Foldit puzzle in the last five years that used the Exploration Condition, this is the same feature.

In the early days of Foldit, there was a completely different type of "Diversification" puzzle that used a different method to measure solution diversity. In those puzzles, diversity was measured between all player solutions (not just a pair of structures). That type of puzzle has been deprecated and is no longer in use. If you're a newer player and don't remember Diversification Puzzles, then don't worry about it—that method was more complicated than the RMSD metric, and I don't think it will help you understand the current Partition Puzzles. However, if you're curious about some Foldit history, the Diversification Puzzles are explained in this 2011 blog post.

jeff101's picture
User offline. Last seen 45 min 8 sec ago. Offline
Joined: 04/20/2012
Groups: Go Science
Would an RMSD-dependent bonus help?

I wonder if these puzzles would be easier
if they had a score bonus that depended
on the RMSD from the starting structure.
This score bonus could be as follows:

RMSD*400 points for RMSD<2.5 Angstroms
1000 points for RMSD>=2.5 Angstroms

This type of bonus would help Recipes
gradually convert structures with small RMSD
to ones with RMSD over 2.5 Angstroms.

You could keep the condition that to receive
credit, structures must have RMSD over
2.5 Angstroms.

You could also remove the RMSD bonus
from the scores when making the partition
functions for these puzzles.

rmoretti's picture
User offline. Last seen 2 hours 29 sec ago. Offline
Joined: 01/15/2010
Groups: None
Might not help much

It's a good thought, but chances are that those solutions which are going to be truly effective in "winning" these second round partition function challenges aren't likely to be the ones which just meet the 2.5 Angstrom cut-off. Remember, the goal here isn't simply to get the highest energy structure or the structure closest to the native. The goal in this second "challenge" round is to come up with structures which "break" the partition function/folding funnel of the design. (Or rather, to point out the existing flaws of the folding funnels of the designs.) Structures close to the design aren't unlikely to be quite as effective in "breaking" the partition function as ones which are further from the design structure, and the more structures you can come up with that differ significantly both from the input and from each other, the more you're able to "break" the partition function.

We can't say for certain, but chances are the people with the most success in these puzzles aren't going to be those who just make minor alterations to the input structure. Instead, I'm guessing that the people with the most success in this round will likely be those who pull the protein apart and then refold it into new and different structures/topologies.

That's sort of the reason for including the RMSD filter in the first place. We don't want you spending a lot of time/effort in structures which are more-or-less the same as the design. These aren't going to be successful in achieving the goal of this second round.

jeff101's picture
User offline. Last seen 45 min 8 sec ago. Offline
Joined: 04/20/2012
Groups: Go Science
Would an RMSD-dependent bonus hurt anything?

Perhaps an RMSD-dependent bonus would
help more than it hurts. The 3 conformations
for fiendish_ghoul's design shown at:
https://fold.it/portal/node/2005620 and
https://fold.it/portal/node/2005623
differ mainly in where the 3 helices are
placed with respect to the sheet made of
3 parallel b-strands. This reminds me of
the conformational isomers of butane:
https://en.wikipedia.org/wiki/Conformational_isomerism
which can interconvert by rotation about
a single bond. In butane, 2 methyl groups
are moving around a covalent bond. In
fiendish_ghoul's design, 3 helices are
moving around a 3-stranded sheet.
Perhaps the RMSD-dependent bonus would
be enough to help a protein convert from
one conformation to another, while keeping
much of the structure intact. If nothing
else, an RMSD-dependent bonus would
give another, more automated, avenue
for exploring the conformations and
energy landscape of each given protein.

jeff101's picture
User offline. Last seen 45 min 8 sec ago. Offline
Joined: 04/20/2012
Groups: Go Science
RMSD-dependent bonus also acts like a shield:
If you included an RMSD-dependent bonus,
I think it would also act like a shield, 
keeping recipes from exploring the range 
where RMSD < 2.5 Angstroms (I call this
range the "RMSD 0 basin" below).

Say you started a puzzle with the protein 
chain linear & fully extended. Odds are 
this would have RMSD > 2.5, but the overall 
score would be very low. Using wiggle or 
just about any recipe, the score would 
gradually rise, and the RMSD would likely 
change. If the designed structure with 
RMSD = 0 has a large basin of attraction 
or large funnel in the energy landscape, 
many folding pathways will eventually 
fall into this basin or funnel. Thus, 
as your score rises, the RMSD value will 
gradually decrease and might drop below 
2.5. When the RMSD drops below 2.5, one 
could say the structure has fallen into 
the RMSD 0 basin.

With no RMSD-dependent bonus, there is 
no penalty to recipes for continuing 
deeper into the RMSD 0 basin. A player 
must notice what has happened ("my score 
has a red line through it and I am no 
longer getting credit for it!") and 
take manual corrective actions.

With an RMSD-dependent bonus, structures 
falling into the RMSD 0 basin will face 
a drop in score, and this drop may be 
enough to redirect recipes away from 
the RMSD 0 basin.
jeff101's picture
User offline. Last seen 45 min 8 sec ago. Offline
Joined: 04/20/2012
Groups: Go Science
RMSD-dependent bonus with a long tail:
A long-tailed form for the 
RMSD-dependent bonus could be:

bonus=1000/(1+(2.5/RMSD))

which gives the bonuses below:

RMSD    bonus
-------------
 0          0
 0.3125   111
 0.625    200
 0.8333   250
-------------
 1.25     333
 1.6667   400
 2.5      500
 3.75     600
 5        667
-------------
 7.5      750
10        800
20        889
huge     1000

As you can see, this bonus rises fast for 
RMSD<2.5 and more slowly for RMSD>2.5. 
The long-tail at large RMSD will encourage 
recipes to explore increasingly larger RMSD 
values, even after the RMSD has passed 2.5.
jeff101's picture
User offline. Last seen 45 min 8 sec ago. Offline
Joined: 04/20/2012
Groups: Go Science
More Systematic Puzzle Names:

Would it be easier if these puzzles had numbers?
The original Phase Two Partition Function Puzzles
could have numbers like 1575a-t, with the 20 letters
a-t being a shorthand for each designer's player
name. The revised versions (B right now) of these
puzzles could have numbers like 1576a-t. Also,
if an individual puzzle, like 1576d, needs to be
revised, it could be reposted with numbers like
1576d2 1576d3 etc.

Of course, if any puzzle gets revised and reposted,
it would be nice if a notice like 'Puzzle Expired'
would appear in the top-center Rank/Score box of
any Foldit client running an old version of the puzzle.
It would also be nice if we could upload our solutions
from previous versions of a puzzle into the latest
version of that puzzle.

Thanks!
Jeff

Joined: 09/24/2012
Groups: Go Science
Scripts

It seems that scripts with "Are conditions met" option don't act as expected.

Is it a bug ? or should we adapt most recipes with a specific filter condition ? ;P

Joined: 09/24/2012
Groups: Go Science
Target

Am I correct if I set, as a target, the score of the starting pose?

If I pass this, I win the battle?

bkoep's picture
User offline. Last seen 1 hour 52 min ago. Offline
Joined: 11/15/2012
Groups: None
Not exactly

You should aim to get the highest score you can. The higher your score, the more you hurt the partition function of the design. It doesn't matter so much whether you pass the score of the starting pose. For most of these puzzles, we don't expect anyone to pass the starting score.

If you do manage to pass the score of the starting pose, that means you have found a decoy that is more stable than the starting pose. This would be really bad news for the design, and that design would not definitely not win the tournament (and in that case I would recommend you turn your attention to another partition puzzle). But that doesn't necessarily mean you "win" the partition puzzle. After all, someone else could find another decoy in the same puzzle that scores even better.

Joined: 09/24/2012
Groups: Go Science
When to stop a partition slice and start another one?

Can you answer the question by susume here:
https://fold.it/portal/node/2005660#comment-37539

According to what susume observed, if any player reaches about 30 pts from the stating pose, it'd be enough to say that the puzzle isn't very stable. This would be a signal to other players to forget this puzzle and to try contesting other puzzles. Alternatively, it can be interesting for science to try another pose for the same puzzle (in order to fulfill other slices of its partition function).

Sorry for the dude questions and thanks for your patience.

alcor29's picture
User offline. Last seen 6 hours 13 min ago. Offline
Joined: 11/16/2012
Is the effort justified.

If we can't get within 100 energy points of the original given that
[E=(F-8000)/-10], then is there a benefit? If we can't get within 50 energy points?

Get Started: Download
  Windows    OSX    Linux  
Windows
(7/8/10)
OSX
(10.7 or later)
Linux
(64-bit)

Are you new to Foldit? Click here.

Are you a student? Click here.

Are you an educator? Click here.
Search
Only search fold.it
Recommend Foldit
User login
Soloists
Evolvers
Groups
Topics
Top New Users
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