Hi everyone!
We've just posted an update to the developer preview which gives you access to a different kind of exploration puzzle.
We've attempted to address many of your concerns from the first exploration puzzle type. In this version, you get score from protein stability like in normal puzzles, but you gain an exploration multiplier for exploring away from the starting structure.
You can play the puzzle by joining the developer preview and then selecting Exploration Puzzle Version 2 from the puzzle select menu.
Your feedback on this new system would be great!
Q: Where can I find the Foldit data files?
A: The data files are in different places depending on your OS:
- on Windows XP at C:\Documents and Settings\All Users\Application Data\Foldit\
- on Windows 7 and Vista at C:\ProgramData\Foldit\
- on a Mac, Ctrl-click on the app and choose "Show Package Contents, then go to Contents/Resources
- on Linux, is is the Foldit directory
Q: How do I try out the Foldit Developer Preview?
A: Locate your Foldit data directory (see previous question). Open the file options.txt and change the update_group variable from "main" to "beta".
See attached screenshot. To discourage such a tactic, it might be a good idea to take the square root of the multiplier.
As far as the giant helix exploit, we know about this and are working on a fix. Taking the square root of the multiplier wouldn't really solve the issue, as it would equally penalize legitimate exploration.
As far as LUA functions,
get_score now always returns the normal "score" (aka stability)
get_exploration_score will return the exploration score (in the old style exploration puzzles) or the exploration multiplier (in the new style exploration puzzles).
get_ranked_score will return the score which you are actually being ranked with.
The rationale behind my idea of taking the square root of the multiplier is really to introduce diminishing returns; exploration is not being penalized, just rewarded less. The intent is really to find a way to prevent the giant helix exploit from becoming a dominant strategy, for which even taking the square root of the multiplier seems insufficient:
In this puzzle, for example, the 9533 x 2.98 = 28383 score I've shown in the screen shot can only be matched via scores such as 28383 x 1.00 or 14191.5 x 2.00 with the current scoring system, but neither seems realistic. Taking the square root of the multiplier means that 9533 x sqrt(2.98) = 16456.5 can be matched by either 16456.5 x sqrt(1.00) or 11636.5 x sqrt(2.00), but these segment scores still seem impractical for a protein this size.
==> Now, there are other safeguards available. Segment score cutoffs would be a good start, but as the last exploration puzzle demonstrates, the all-beta sheet exploit is available when the giant helix doesn't work. One could also manually disqualify giant helices and beta sheets, of course, but this runs the risk of discarding a legitimate solution by mistake. So, I would suggest considering additional options that could weigh the segment score (SS) and old-style exploration score (ES) equally, possibly with diminishing returns factored in.
The current system already gives diminishing returns in the exploration multiplier. The multiplier is a logistic function of the (old) exploration score. So as you get nearer and nearer to 3, it becomes harder and harder to increase your multiplier.
This does not, however, address the issue of the giant helix. All exploration suffers these diminishing returns equally, so the problem persists.
The problem stems from the fact that a giant helix IS much different from the starting structure, and also is "fairly" stable.
A solution we will probably end up using, at least for now, is to not reward contacts from helices (or reward them less). You're right in that this does hinder legitimate exploration to an extent, but it shouldn't be much of an issue. Most of the desired changes are not reshaping segments into helices.
Since you're rewarding exploration in terms of contacts, not in terms of changing predicted secondary structures, how about a penalty for changing predicted (or initial) SS's?
You could make the penalty lighter when the predicted SS's are less certain, so changing structures would be fine if it has basis in actual data on the protein.
You could also allow a certain percentage of "free" SS changes to keep the spirit of exploration, but make the penalty get stiffer with each additional change after a certain point.
I think this discourages the giant helix better than the options I've seen so far without being arbitrary.
That's exactly what the solution I proposed does. It is in no way arbitrary.
Helices show up on the contact map as a very regular and predictable pattern (they are contacts that run in a line along the diagonal of the map).
If you have a helix, you'll have a line of these contacts. If you don't have a helix, you wont have a line of these contacts. There is no way to decouple them.
If we don't award players for contacts in this area, it is no longer beneficial to try and make helices.
The only caveat to this is that these contacts sometimes are part of other structures as well, such as very tight loops. But in those cases, you're missing out on only a few contacts, so the impact is minimal. Trying to prevent helices in a different manner would likely have much worse side effects.
I apologize – I misread your proposed solution in comment #6. For some reason I thought you were referring to contacts between helices and other structures, rather than contacts within helices. So I agree that you've addressed the helix issue.
As infjamc mentions, there may still be an issue with sheets, and that was what I was trying to address with my proposal (in addition to helices). The hard part would presumably be detecting sheets structurally, to deal with the tactic of reassigning SS types after creating multiple sheets.
The contact map doesn't help in this case, but is the "zig-zag" pattern of a well-formed sheet enough to distinguish it from a loop in the folded protein? It sounds tricky to implement, but if you really want to avoid game-breakers, it may be necessary to do something like this eventually.
We are actually interested in exploration in terms of sheets, so this shouldn't be a problem.
Generally, I think we usually have a good idea of where the beta strands are in the protein, but not how the sheets form out of these strands. So things like register shifts or forming new sheets are (and should be) fair game.
Yes, players could try to create sheets out of everything, but it I don't think it will be an issue.
The issue with the helix strategy is that it was very easy to do and netted a very high score. Getting a high score by trying to form everything into a sheet is arguably harder. I'm sure someone will try it at some point, but for the most part I would expect that people will decide this approach isn't worth their time and will continue exploring legitimately.
