1667: "Too Much Density!" Freestyle puzzle

Closed since almost 7 years ago

Summary

• Created: April 25, 2019
• Expires: May 06, 2019 at 07:00 UTC
• Max points: 100

Description

This Electron Density puzzle is different than anything we've ever posted before! The extended chain is 187 residues long, but we are giving you a lot more density than that. You'll need to find where to fit this chain in the giant blob of density. (The "Trim Density" tool will be your friend)

Top groups

• 1. Go Science 100 pts. 22,258
• 2. Beta Folders 73 pts. 20,387
• 3. Anthropic Dreams 52 pts. 20,350
• 4. Gargleblasters 36 pts. 20,245
• 5. Russian team 24 pts. 20,076
• 6. Hold My Beer 16 pts. 20,019
• 7. Void Crushers 10 pts. 19,955
• 8. Marvin's bunch 6 pts. 19,897
• 9. HMT heritage 4 pts. 19,876
• 10. L'Alliance Francophone 2 pts. 19,761

Top soloists

• 1. fiendish_ghoul 100 pts. 22,033
• 2. Bautho 97 pts. 21,534
• 3. Bruno Kestemont 94 pts. 21,104
• 4. Galaxie 91 pts. 20,343
• 5. johnmitch 88 pts. 20,331
• 6. LociOiling 85 pts. 20,264
• 7. pvc78 82 pts. 20,240
• 8. Deleted player pts. 20,239
• 9. Susume 77 pts. 20,203
• 10. vakobo 74 pts. 20,076

Comments

[DoctorSockrates]

Hey folks, your favorite young streamer's back with some preliminary comments and it's…not fantastic to say the least.

I actually had a lot of fun with the last couple ED puzzles because my strategy would be to get a lock on a region with a sequence of "landmark sidechains" (namely the ones with benzene rings) and proceed to attach them one at a time from there. Quite therapeutic once I got "into the zone" and got a steady flow going.

But then I got to this one. For those keeping up with my livestreams you probably saw last Friday that I wasn't able to get a lock on any portion of the cloud for a couple reasons. Firstly was discovering how Trim Density actually works.

What I expected it to do was cut away a portion of the cloud based on the slider from the point of reference of the camera. That probably sounded wordy, so I'll provide an example: If I looked at the cloud from afar, slid the slider to 75% and hit okay, I'd expect the 25% closest to the camera to cut away, exposing the layer underneath.

How it actually works is that it cuts away all of the cloud except for any bits that are close to the physical protein that you're manipulating, the distance threshold obviously being modified by the slider. So for those like me wondering why the entire thing disappears when you hit OK, this is why. Now I tried using it the proper way once I discovered what was wrong…but honestly with how it worked I didn't feel like it was useful. I'd have to spend time folding up a protein into a rough chunk of what I wanted to focus in on, resize it, re-fold it into the residues that I wanted to test for "landmark" status, rinse repeat. That's way too much effort for trying to insert a residue like a literal puzzle piece to see if it fits the metaphorical "hole" that's obscured by the rest of the holes (i.e., the rest of the cloud.)

Electron density puzzles are otherwise simple to me conceptually: find the patterns in the cloud, match them with what you have, and align accordingly. What strikes me as the main difficulty here is the workspace. Threshold manipulation was enough for me to scope out the landmark residues on the outside for me to align the rest, but once you layer in the shell of extraneous cloud, it's not impossible to do the same strategy, but it's much too time consuming to hold my attention in its current state.

I've seen frood's eraser suggestion, and that could be a good remedy, and it's quite similar to a solution I had in my mind where one could cut out fragments like a surgeon to expose the cloud underneath. Controls have always been a bugbear with Foldit and if I were in the developer hotseat I'd be looking at other 3D modeling programs or sandbox games as a reference to improve the UI/UX shortcomings of Foldit.

I'm gonna separate this post so that it doesn't get too long, so check the replies.

[DoctorSockrates]

So another thing that is probably holding me back is that I prefer to approach from a gameplay-first, reference later perspective. As in: I don't look up the puzzle sequence on the PDB. All power to those who do, but for me, I have reasons.

I don't necessarily do this with other games. I've spent more hours in the Runescape, Minecraft, Terraria, and Warframe wikis than I care to admit. But with Foldit my personal justification (and emphasis on personal; people should have every right to play the way they want) is that with Foldit, I should ideally be relying on the patterns I see given the game data in front of me to play, to an extent. It's also worth noting in those other games that the core gameplay loop is engaging and satisfying enough for me to justify looking at pages of words and documentation to get the most out of my time there, whereas in Foldit, I'd feel like I'm missing the point if I go to look at references. Not to discount the collaborative nature of the game, of course, but this is how I get myself to practice improvement.

So, for example:

• In Prediction puzzles like Unsolved DNFs, I'll make little changes to the pre-set PSIPRED secondary structures and work with those for as long as possible to put together my rough draft fold
• In Design puzzles I often do not consult the PDB for inspiration on what I should make; I simply kitbash together building blocks at my leisure in the hopes that it creates something pleasing and roll with it. I follow the rules of the game of course, but otherwise my designs are rather random!
• In Revisiting puzzles the only hints I allow myself are the proper disulfide bridges when applicable. I otherwise again work from what I'm given.

So for this puzzle, I tried to employ my strategy for finding the landmark residues outlined in my previous post, and had difficulty. This was the one time I actually broke tradition and ran PSIPRED on the puzzle sequence (which I had to use the print protein recipe for) just to try and find an extra clue. During stream you could see me try to no avail to align the predicted secondary structures.

This brings me to the puzzle design. Foldit's honestly the hardest to do, because you're pitting the difficulty of actual science with the entertainment value of a game. I really don't want to go looking stuff up only to find that having the knowledge didn't really help my gameplay experience to much (or compromised the sort of "point" of the game to come up with predictions ex nihilo.)

Given the current state of the controls, the difficulty of the puzzle, and the time investment needed to come up with a solution, this puzzle is poorly designed from a gameplay perspective. From a science perspective, I actually think that it's ideally designed! But as with a vast majority if not all scientific studies: High internal validity usually means lower external validity. Bridging the analogy here: This puzzle has rock solid integrity in terms of scientific practices and theoretical application (I love the idea of this being the hardcore standard of bringing the game close to what actual microscopists have to deal with), but as a difficult puzzle for a video game it's pretty much restricted to any savant who has the time, energy, and force of will to take on these challenges willingly. And that, I assure you, is a minority for any video game.

[DoctorSockrates]

Mind you this is coming from an end-user perspective, but is it just me or is half of the threshold slider useless? Legit, the entire left side of the slider results in too noisy of a cloud to use.

I feel like that entire left side of the slider can just be cut out and the other half expanded to the original scale so that we could have finer control of what we want to see.

Even better, add numerical input boxes to all sliders.

If anything this puzzle is just another reminder of how much we take for granted when it comes to UI/UX polish. Sometimes nobody thinks to add a feature until someone brings it up, and that's why playtesting/feedback response is so important in games. Of course, I'm not going to ask the impossible; developer resources and working time are finite, and the last thing I want is someone to crunch in order to make our lives as players marginally better. But, as I have heard from accessibility advocates in gaming, if Accessibility/Quality of Life features is/are planned out from the get-go or sooner rather than later, everyone benefits (see the "Curb Cut" effect for a deeper justification).

[jeff101]

One recipe I've been using for ED puzzles is called BandOntoSphere1.0
(https://fold.it/portal/recipe/102961). I made it to keep the protein inside
a sphere centered on the center of the ED cloud. Before running this recipe,
I identify a point in space roughly at the center of the ED cloud and make a
band from a segment of the protein to this point in space. This band is called
the hub band. Then, when I run the recipe, it makes a bunch of bands from
evenly-spaced segments of the protein to the center of the ED cloud based
on the hub band. I set keephub to 0 to remove the hub band before the recipe
finishes. For Puzzle 1667, I have found that setting the band length to 36 and
the band strength to 0.1 work well. Otherwise, I use the default settings.

If the band strength is large, BandOntoSphere1.0 forces the protein to lie on a
thin spherical shell with radius equal to the band length setting. With lower
band strength, the thickness of the spherical shell increases. Setting the
band strength to 0.1 lets the protein fill a good portion of the sphere.

This recipe would work better if Foldit had LUA commands for one-way bands,
in particular, the type that keeps the band length shorter than a specified value
(https://fold.it/portal/node/996372#comment-27202). This kind of band
would be enabled when the band length is above a certain value and
disabled when the band length is below a certain value.

[frood66]

I agree BP - but that is the way this place has gone.

With regard to this puzzle - the definition is too poor to do much else…..

But - on this occasion the native is mostly not gonna score well (other than density) which brings us back to whether it's all worth it.

So the game is being hoist by the petard of the setters.

It's a shame.

It would all make more sense if the cloud definition was better - imho.

[frood66]

I think a trick has been missed here.

If this puzzle had been set with better definition (this one is real poor) I think many would have embraced and enjoyed the challenge.

But it has not.

Net result - few can even be bothered.

I'm a true exponent for ED - but this is not set up to enthuse.

Suggestion for FC - if U R going to create a huge puzzle with much cloud that is extraneous - at least
allow for a level of definition that does not cause despair.

It's bad enough that the memory usage is massive - but to not even throw a bone to yr dogs is daft.

[beta_helix]

Thank you all for your candid and specific feedback about this puzzle.

With the success of the previous group of cryo-EM puzzles, we are trying to figure out how to present future puzzles where the density has not been successfully trimmed by a microscopist.

Clearly, this will require several new Foldit tools in order for this to be feasible… so thank you very much for pointing those out! Some of these will take a lot more time to implement (such as the ED eraser tool), but some tools should be easier to add to the game (such as adding numerical input boxes to all sliders and LUA commands for one-way bands).

We realize this particular protein exists in the PDB, but there is no other way for us to post such puzzles… unless there was a specific CASP for cryo-EM structures (which would be awesome because all of you would do so well!).

Thank you for your patience on this one, we won't post another similar puzzle like this until we come up with a better way.

[jeff101]

In my efforts to identify amino acids in this puzzle's ED cloud,
I asked myself if CO groups on the backbone make any features
(like little spikes or bumps) in the ED cloud. I don't recall
seeing such features in previous ED puzzles, but sidechain CO
groups certainly give features in the ED cloud. In the image at
https://fold.it/portal/node/2001386 sidechain oxygens seem to
act like sidechain carbons within the ED cloud. I also think the
COO group at the COO terminal of a protein chain shows up in the
ED cloud, looking like an extra valine or threonine sidechain there.

So, do backbone CO groups give features in the ED cloud?
If so, what do they look like? If not, why not?

Thanks!

[bkoep]

Unfortunately, I don't think there's any direct way to emphasize the density of sidechains. Cryo-EM only tells us where electrons are concentrated—it doesn't distinguish between electrons on sidechain atoms vs. electrons on backbone atoms.

We could think about calculating something called a difference map, using the player's solution. This means we would calculate a theoretical electron density cloud from the player's model, then subtract the theoretical cloud from the EM cloud. The result would be a "difference cloud" that highlights regions of density that are not occupied by the model.

For example, in a case like this one, where you can pick out helices but can't see the sidechains: the player might build the helices out of GLY and then compute a difference map. In the difference map, the theoretical density of the GLY helix would "cancel-out" the EM density along the helix backbone, and the density for the (missing) sidechains would be relatively more intense.

This wouldn't improve the "quality" of density at the sidechains, but it might be useful for visualizing weak sidechain density.

[horowsah]

The answer is that they do, if you have high enough resolution data to see them stick out (usually need about 2.8 angstrom resolution data or better for that to happen). In this puzzle, the data isn't good enough, which is usually the case with electron microscopy-derived density. This is one of the things that makes building into these maps a challenge.