Closed since almost 14 years ago
Advanced Overall Prediction Electron DensityUp until now, we've been posting contrived electron densities in order to help us refine the software and puzzle parameters. We now feel confident that both players and tools are ready for real-world densities that are generated from experimental data. For this puzzle, we've taken a large protein and broken it up into 3 domains. This puzzle gives you the smallest of the domains, made up of 2 different chains, along with the electron density that was generated from experimental data. As always, try to fit the density - but keep in mind that real world data isn't quite a ideal as previous puzzles, and it's possible it might be wrong or imprecise in areas.
My score drops by a lot whenever I align. Shouldn't the alignment algorithm make the alignment bonus improve when I align the protein to the ED, regardless of whether the ED is correct or not?
(Currently (8698) my OTHER score is -84 vs -157 aligned. I did a reset and wiggle, that is -80 vs -97 aligned, after fuzing -84 vs -109)(minimum threshold maximum cloud while aligning)
Your intuition is correct - aligning should always increase your score.
However, the problem is that aligning isn't easy - we cant just find the perfect answer like we can for aligning a QTTN. When you press the align button, it does a bunch of complicated math to make a best guess of how to orient the protein, and then does a type of wiggle to let it settle into the density. This doesn't always get the right answer, so your score can actually drop. If you already have the protein well aligned within the density, pressing the "Refine Alignment" (I think that's what it's called…) will JUST perform the wiggle part, and will only ever increase your score.
It might make sense to have the Align button only actually replace your current alignment if the score you end up with is better. I can try this out and put it in if it works well.
Until this puzzle, you are right, jflat, the "Refine" has always increased score - but on this one, I'm seeing it drop by fractions instead of increase.
???
Forgive me, I thought this was referring to the other button. It does seem that refining will lead to a drop in score in some cases, which should not be happening. I'll look into this.
Alright, I've discovered why aligning is resulting a score decrease -
The problem is that when you press align, foldit tries to orient the protein in the way such that it gives you the most ED alignment score. Now, for a protein with a single chain, this works great. But this puzzle has two pieces, and it is trying to fit each piece into the density separately. So the relative position of the two pieces is changing when you press align. This is somewhat desirable, since you are matching the density better.
However, I think it is still better for align to only align, so that it can never decrease your score.
