Placeholder image of a protein
Icon representing a puzzle

775: Player-Solution Redesign: Negative Design

Closed since about 9 years ago

Advanced Overall Design


September 06, 2013
Max points

In this puzzle you can redesign a player solution from Puzzle 742. The starting structure is the monomer subunit for a top-scoring solution by Galaxie, jamiexq, and tokens. See if you can find a higher-scoring sequence that stabilizes the design fold! This puzzle has an RMSD condition, so don't stray too far from the starting fold. We've also included alignments to two decoy folds—alternative high-scoring folds for the same amino acid sequence. New designs that also destabilize the decoy folds will be prime candidates for synthesis in the Baker lab! See the puzzle comments for more details.

Top groups

  1. Avatar for Russian team 100 pts. 9,956
  2. Avatar for Gargleblasters 2. Gargleblasters 80 pts. 9,917
  3. Avatar for Beta Folders 3. Beta Folders 63 pts. 9,898
  4. Avatar for Anthropic Dreams 4. Anthropic Dreams 49 pts. 9,840
  5. Avatar for Contenders 5. Contenders 37 pts. 9,831
  6. Avatar for Void Crushers 6. Void Crushers 28 pts. 9,819
  7. Avatar for L'Alliance Francophone 7. L'Alliance Francophone 21 pts. 9,810
  8. Avatar for Go Science 8. Go Science 15 pts. 9,788
  9. Avatar for Deleted group 9. Deleted group pts. 9,764
  10. Avatar for SETI.Germany 10. SETI.Germany 8 pts. 9,751

  1. Avatar for Grom
    1. Grom Lv 1
    100 pts. 9,912
  2. Avatar for cbwest 2. cbwest Lv 1 99 pts. 9,892
  3. Avatar for pvc78 3. pvc78 Lv 1 97 pts. 9,866
  4. Avatar for hpaege 4. hpaege Lv 1 95 pts. 9,861
  5. Avatar for Sadler 5. Sadler Lv 1 93 pts. 9,858
  6. Avatar for deLaCeiba 6. deLaCeiba Lv 1 92 pts. 9,855
  7. Avatar for silverberg 7. silverberg Lv 1 90 pts. 9,848
  8. Avatar for decbin 8. decbin Lv 1 88 pts. 9,843
  9. Avatar for Galaxie 9. Galaxie Lv 1 87 pts. 9,840
  10. Avatar for vakobo 10. vakobo Lv 1 85 pts. 9,835


bkoep Staff Lv 1

After Puzzle 742 closed, scientists at the Baker lab flagged this solution as a promising design, and we sent the design sequence to Rosetta@home for ab initio folding simulations. Those simulations produced hundreds of thousands of structure predictions; each prediction is represented in the graph below as a single red dot, according to its potential energy ('score') and its dissimilarity to the design ('rmsd').

The good thing is that the predictions with the lowest potential energy are within 2 Å RMSD of the design (i.e. the most stable structure predictions are very similar to the design structure). However, we're a little worried about the two funnels on the right side of the plot, which represent 'decoy' structure predictions. Those structures have very different folds from the design although they are nearly as stable.

Before we try to make this protein in the lab, we want to be sure that the design structure is the most stable fold for the design sequence. This is our first attempt in Foldit towards what we call "negative design." We want to see if you can widen the energy gap between the design structure and these two decoys, by making mutations that destabilize the decoy folds.

gitwut Lv 1

IANAMB, and am somewhat dense too. :)

What are we supposed to do with this puzzle? At first it seemed simple:

See if you can find a higher-scoring sequence that stabilizes the design fold! This puzzle has an RMSD condition, so don't stray too far from the starting fold. Then you add decoy alignments to the mix.

What are we supposed to do with the two decoy alignments?

Are we supposed to overlay our above stabilized protein with each of the alignments and see what happens?

Are we supposed to find the highest scoring solution for each of the alignments?

marie_s Lv 1

If I understant the goal is to make a sequence with big score in tmpl1 and bad score in tmpl2 and tmpl3 after shaking.

tokens Lv 1

Another question: As this is supposed to be a trimer, how do you make sure that we don't destroy the interface to the other two subunits when we start mutating? With only one subunit present in the puzzle, it seems likely that our mutations could remove the sidechains creating this interface.

bkoep Staff Lv 1

First, I'll clarify that the decoy alignments play no role in the scoring. So if you are just playing for the best rank on this puzzle, you can ignore the decoy alignments. However, if you get bored with that, or if you like the extra challenge, you can try to make mutations that cause the other alignments score badly.

marie_s is correct in her comment below. If you align the starting structure to any of the templates, then shake and wiggle, you'll notice that they all have similar scores. You want a design that scores well when aligned to TMPL1, and that scores poorly when aligned to TMPL2 or TMPL3. A simple and effective approach to checking your progress could be to: align to the decoy, shake and wiggle, compare score to design.

bkoep Staff Lv 1

(Oops: The alignments were meant to have more descriptive titles, but I must have overlooked this when posting the puzzle. TMPL1 is the design; TMPL 2 and 3 are the decoys. Apologies for any confusion…)

bkoep Staff Lv 1

Good question. In fact, I would have liked to post this puzzle as a trimer, but there were some technical problems (read: Foldit bug) using the alignment tool with a symmetric puzzle. Rather than wait for another update to fix the issue, we opted to try the negative design puzzle for the monomer. Even if you destroy the trimer interface, a successful negative design puzzle would be pretty cool in and of itself!

utaca Lv 1

It is not possible to use banding scripts like ravens BiS Flu because the exploration score gets immediatly higher than 2.0 and the score is lined out. Same issue if you move the protein with the pink cross (moving tool?). It's difficult to get the protein back on the grey line.
How can I align the protein back to the grey "exploration line"?

utaca Lv 1

Thanks silver and brow!

After playing a little around i found that it's the best to minimize from time to time the exploration number by using the pink cross.
For raven's GAB BiS I found the best lined-out score in cntrl-3 and moving back by handwork. But does every script save a lined-out score?

Why shouldn't I get the points when the protein is only a little displaced? I understand that the devs need a tool to determine if the the fold is similar to the original but the grey line as a part of the score is too static.

Happy folding


P.S.:Another trick I tried didn't work: Loading the reset version as a guide and then align to guide.