Closed since over 12 years ago
Advanced Overall Prediction Electron DensityThis unsolved bacteriophage protein is the same protein that you've been working on in puzzles 737, 740, 749, and 756. We've analyzed the top solutions from the latest puzzle, in which you tried to fold an extended chain into an electron density cloud, and we want to see if you can keep improving on this structure. This puzzle includes the three most promising Foldit solutions (by Galaxie, meatexplosion, and karstenw) as starting structures, and the electron density cloud is a little more coarse. We're also providing you with a structure that was determined by another lab (PDB entry 3j40), although we're fairly certain this solution is inaccurate. You can load in your solutions from the previous puzzles (737, 740, 749, and 756) and can find more details in the puzzle comments.
100 pts.
12,105
NOTE: If you did not manually save a solution that you want to use from puzzles (737, 740, 749, or 756), you can go back to those puzzles, manually save it, and the solution should appear in your manual saves for this puzzle.
Original secondary structure predictions from the SAM server are here:
http://fold.it/portal/node/995398#comment-23593
More info about Cryo-electron microscopy here:
This has made seeing features in the cloud a lot harder. Dont like it.
We realize the coarser density map in this puzzle makes things more difficult—I should explain the reason for that.
The original density map we have is an average density map, constructed out of many, very noisy, data from cryo-EM experiments. Most features of the average map are real, but other features are false artifacts. So even if you could place an atom perfectly into every feature in this map, some of those atoms would be in the wrong place.
To control for this "over-fitting" to the density map, we split the data into two sets. This way, you can use data from one set to guide your predictions, and we can use data from the other set to validate your predictions. If your solution matches the native protein, then it should fit both maps equally well; if your solution is "over-fit" to your half of the data, then the solution may fit very well into your prediction map, but will fit our validation map very poorly.
Unfortunately, when we use only half the data to calculate the new average, the resulting map is more noisy. So this map will be more difficult to use, but we have a way to tell how accurate your solutions really are.
The cloud in 756 was not "sticky" enough - not enough score boost, so the protein would not stay in it but would wander around instead. It also had a lack of recognizable landmarks (sidechains). The current cloud I find not useful at all.
This iteration of the cloud is more annoying than helpful, at least for play. Can't see where anything should go, no landmark sidechains or bonds - it's just obscuring the view and not guiding the fold.
I would say: repost the puzzle with the original cloud. (Exactly the same so earlier solutions go to the same position).
And if you want with some player solutions.
I am convinced I found a good starting point the last 2 days but ran out of time to fully explore it. And with this cloud I cant use that solution.
Hey bkoep,
In cryo-EM is it customary to split the data 50/50 into working and test sets? This is the treatment you implied above. or is it sufficient to split it 90/10 or 95/5? and would that give a more meaningful map?
Good question, Anfinsen. With cryo-EM data, we're limited in how much we can reduce the size of the test data because other sources of noise become more prominent (namely, particle orientation error). I'm afraid that if we altered the distribution further, the testing map would be useless before we could see any appreciable improvement in the working map.
