[Devprev] Density Refinement Puzzle

Closed since about 2 years ago

Summary

• Created: October 26, 2023
• Expires: February 05, 2024 at 16:00 UTC
• Max points: 0

Description

With the addition of the Electron Density Panel a few years ago, Foldit unlocked a new way for players to help solve complex protein structures. Our latest update has added a tool that is going to take electron density in Foldit to the next level - the Refine Density button.

Check out our upcoming blog post to see exactly why 'refining' (a.k.a 'rephasing') density is such a powerful new addition.

NOTE: Reopened for debugging purposes!

Top groups

• 1. Anthropic Dreams 1 pt. 28,538
• 2. Go Science 1 pt. 28,352
• 3. Contenders 1 pt. 28,328
• 4. Gargleblasters 1 pt. 28,120
• 5. L'Alliance Francophone 1 pt. 28,023
• 6. Firesign 1 pt. 27,141

Top soloists

• 1. LociOiling 1 pt. 28,538
• 2. blazegeek 1 pt. 28,457
• 3. akaaka 1 pt. 28,392
• 4. Artoria2e5 1 pt. 28,352
• 5. BootsMcGraw 1 pt. 28,328
• 6. Aubade01 1 pt. 28,182
• 7. Bletchley Park 1 pt. 28,143
• 8. HuubR 1 pt. 28,120
• 9. rosie4loop 1 pt. 28,079
• 10. apetrides 1 pt. 28,068

Comments

[LociOiling]

Saved solutions may load lower when you open them on a different client.

The updated density map you get using the "Refine Density" button is apparently not part of a saved solution.

Clicking "Refine Density" on the new client seems to find the missing points, at least in one quick test.

Evolvers will probably need to use "Refine Density" on solutions which previously used the tool. Likewise, if you run multiple Foldit clients, you'll need to run "Refine Density" again if you open a solution on a different client.

[LociOiling]

I see this puzzle offers 100 points like a regular puzzle. That seems unusual, most puzzles of this type offer no points.

[rosie4loop]

Quick comments after initial inspection of this puzzle:

Comparing puzzle 2371 and this test puzzle, a ".sf" file is loaded to replace ".odensity", which should stand for "structure factors". The ".sf" file is not encrypted, seems to be in .cif format as I inspect it in a text editor. Information from this file also gives the PDB ID of the density map:

_entry.id   7tvc 

In previous ED puzzles, the map is trimmed around the structure. In this puzzle, the whole map is loaded, to allow density refinement after modifying atomic coordinates. Hence the tool of "trim density" or reducing the radius of the density viewer to show only density around the focus would be very useful in this puzzle.

Other remarks and quick questions
It is really nice that we can finally refine the density map, making it more like practical structure solution.
I wonder is there any plan to let us load the difference map in Foldit, so the map is more informative than a single 2Fo-Fc map?

(edit: remarks on trimming density or changing radius around focus)

[rosie4loop]

After a density refinement, clicking "restore very best" only loads the score and coordinates, but not the density. a subsequent wiggle optimize against the updated density and may lead to a lower score.

should be the same issue as the first comment of this puzzle.

Restarting the client also recalculates the map. Which means, if you modify the structure and close the client without refining density, restarting the client will get you a differnet score.

[rosie4loop]

2Fo-Fc map is model biased, which means "maps will tend to show electron density for an atom in the model whether it is truly there or not" but ideally putting an atom in the wrong position should give "only little to no density". (see reference 2).

I believe the score should also reflect the incorrect placement after map refinement. However scoring in Foldit is real-time, which mean the model can be scored with an old map.

Maybe in this kind of puzzles, the score should only be recorded or submitted after the density is refined on the current structure, so the score could better reflect the quality of the model built by players.

Reference:

1. https://proteopedia.org/wiki/index.php/Electron_density_maps
2. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5051661/

[apetrides]

@rosie4loop just want to make sure i understand your feedback when it comes to 'trim density' here- the tool should always be creating a density map that encompasses the entirety of the model. is this not the case in your client?

[rosie4loop]

@apetrides My bad in the poor organization of the post, it was just a feedback on "how to play it" compare to previous ED puzzles.

What I meant was in previous ED puzzles, the map was trimmed before setting up the puzzle, resulting in problems like in https://fold.it/forum/bugs/puzzle-2323-issues-with-missing-densities-in-alternate-locations-chain-a-and-false-residue-connectivity-chain-c

In this puzzle, the density of the whole unit cell would be loaded, which is more useful for handling cases like in puzzle 2323. The downside is that displaying the density of the whole unit cell make it difficult to work on the model, so either the trim density tool or the "radius around focus" tool would be very useful in future ED refinement puzzles.

[BootsMcGraw]

Clicked on "refine density". The first three times, nothing happened. The fourth, it lowered my score, and I couldn't get my high score back.

Kill it with fire. Please don't foist this useless modification upon a dwindling player population.

[rosie4loop]

Suppose it is more "scientifically correct" than the current way of how ED puzzles work, if the score would sync only with the matching model+density, and if the density state could be saved with the model for sharing.

Better disable the score if the model + density does not match. Like after inserting a cut in model, we need to close the cut to get a score. Or like after using the trim tool, we need to untrim to score. Do the same for density refinement puzzle, only after the model+density are in sync the score would be valid.

Previously, the absence of a density refinement tool was one of the major reasons preventing me to use Foldit in training sessions with my crystallographer collaborator, that the current approach with a "single map" is scoring "only on a map with model bias", "in some cases modelling on bias/noise" etc.

[rosie4loop]

Clicked on "refine density". The first three times, nothing happened. The fourth, it lowered my score, and I couldn't get my high score back.

As far as I know the refine density button calculates the map on the current model, so if they're in sync nothing would change.

The score is real-time, though, so it can score the model on the "wrong" map, giving a "wrong" score. I believe the devs better change this behavior to make sure the score better represent model quality.