11 replies [Last post]
Joined: 05/19/2017
Groups: None

Hi folks!

Lately on my weekly livestreams I've been feeling pretty uninspired as far as my monomer design puzzles go, so I was wondering if some of you had anything you'd like to have me try to build in the monomer design puzzles. I do know about the PDB but don't typically have the time to sift through and find something worth replicating that's in line with the constraints of Foldit. If there are any cool proteins to replicate or designs to bring from concept into stable fold, I'm all ears.

That being said, nothing too difficult please; while I know the residue limit already puts a damper on complexity I do like to finish my hand-folding within the course of a stream before "sending it off" to recipe refinement land. I'm also relatively less experienced than most of you all with the game, even though I do feel like I've been improving steadily on the global ladder.

- Sock

spvincent's picture
User offline. Last seen 1 hour 18 min ago. Offline
Joined: 12/07/2007
Groups: Contenders
It's never been quite clear

It's never been quite clear to me what they're looking for in these monomer designs. Exotic folds that rarely seem to score well? Good examples of existing folds? I occasionally see foldit designs in Rosetta@home, and they're usually variations on the hot dog theme so I'd assume the latter. We used to get blog posts and videos on this sort of thing and it would be good to get some more.

(Sorry I know this doesn't really answer the question)

Joined: 05/19/2017
Groups: None
It's a bit too open ended, I think.

This category of puzzles has always been the blank canvas protein sandbox to me. With no constraints or apparent objectives to work towards, I always felt this "mode" was better left as an always available standalone editing sandbox, with no constraints on length. I'm sure this is already discouraged, but what's to stop folks from uploading the same high scoring solution every week the appropriate monomer design puzzle shows up? I feel like having it up as a weekly rotation is more appropriate if there was at least 2 unique, randomly selected constraints in the rotation (e.g. X length + helices/loops only or Y length and at least 1 disulfide), while having the true sandbox mode separate with a perpetual leaderboard.

I guess for me this just highlights a distinct funk in puzzle selection for me lately as a whole. I've been kinda getting complacent with just the cookie cutter usual rotation of revisiting, unsolved de-novo, and monomer design. The only puzzles lately that really got the hamster wheel in my head turning were the small molecule design series, the Aflatoxin Challenge series, and the find the hydrogen bond network puzzles. In all of those I felt like I was working towards a focused goal with nearly infinite possibilities on how to achieve it.

LociOiling's picture
User offline. Last seen 2 hours 54 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders
Helical bundles...new fashion trend?

The question of where we're going with the design puzzles was also raised in chat the other day.

The science team never tells us everything, but there might be some clues to what they're looking for in the recent puzzles.

The current design puzzles don't have don't have secondary structure limits. For a long time, there was a penalty if you had more than half the segments in helixes.

They've also removed the ideal loops and core exists filters (conditions or objectives). There still some restrictions on which amino acids you can use where, and the big rings get some special attention, but it's much more of a protein sandbox than it was for a while.

The surfin' hot dogs that became so familiar will still score well, but the current design rules favor helical bundles.

We've also seen the return of hydrogen bond network puzzles, and these puzzles just happen to involve helical bundles.

In the PDB, there are some examples of designed helical bundles from the Baker Lab, such as 4UOS and 4TQL. They're not exact matches for anything we've seen, but do seem to be somewhat related.

The article High thermodynamic stability of parametrically designed helical bundles from the Baker Lab gives some background on where these proteins came from.

There's also De Novo Design of Helical Bundles as Models for Understanding Protein Folding and Function (*not* from the Baker Lab), with another take on designing with helixes.

Both these papers are a few years old, but the hydrogen bond network puzzles seem to be pointing toward some new activity. Maybe the recent changes to design puzzle objectives are suggesting a parallel track.

One thing that jumps out is that scientists have been successful at getting these lab designs to fold up on their own. Maybe some helical bundles designed by Foldit players will also make the cut. Someday the Design Puzzle Results page on the wiki may get a few more entries.

LociOiling's picture
User offline. Last seen 2 hours 54 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders
More light reading....

There's also this somewhat more recent article that Susume spotted, De novo design of protein homo-oligomers with modular hydrogen-bond network–mediated specificity.

The tease is:

...the approach, which could one day become programmable, [is] to build novel topologies with two concentric rings of helices...

Joined: 05/19/2017
Groups: None
Helices OP?

Reminds me of the time where I managed to get a high score on my Beginner Monomer Design puzzle by helicizing the *entire* peptide. There's gold in them there spirals!

I'll definitely keep this in mind going forward, this is an interesting trend to follow.

bkoep's picture
User is online Online
Joined: 11/15/2012
Groups: Foldit Staff
Think exotic!

Sorry I'm late to the conversation! If you're at a loss as to what you should design, I would encourage you to try building new folds that you haven't seen before, or that you only see rarely.

I think creativity is the biggest advantage that Foldit players have over traditional protein design in scientific labs. In the Baker Lab, I can set up a Rosetta job on a supercomputer that will build and design a thousand 3-helix bundles in a week. What I can't do is tell Rosetta to invent a dozen different protein folds. And even if I sat down and tried to invent a dozen different folds by hand so I could set up a Rosetta job for each (which itself would probably take me a week), all of my inventions are probably going to look pretty similar to the natural proteins that I'm most familiar with.

Foldit players, however, are capable of coming up with wild new ideas that don't quite resemble anything in natural proteins (for example, see this recent Forum post by jeff101). We see this kind of creativity in the Design Puzzle results every week—especially in the Scientist-Shared solutions, which might have a lower Foldit score than a 3-helix bundle at the top of the leaderboard. Unfortunately, the Foldit score is not a perfect indicator of plausibility (this is the focus of recent blog posts, here and here). A lot of these low-scoring Scientist-Shares are physically realistic designs that could very well fold up successfully.

When we upgraded the Foldit score function earlier this year, we dialed back a lot of the filters and objectives that we had been using for protein design in Foldit. One reason for this is that we hoped the new score function would account for some of the filters (like Core Exists) and those filters wouldn't be needed anymore. Another reason is that we were worried that some filters (like Ideal Loops) put too many constraints on the wild things that Foldit players are capable of dreaming up.

However, I think you're right that these design puzzles can be a little too open-ended, and I like your idea of imposing randomized constraints every week. We've been thinking about how to encourage more experimentation and diversity in Foldit player designs, and I think certain kinds of constraints (like the Secondary Structure filter) will be a part of that.

Joined: 06/06/2013
Groups: Gargleblasters
Thanks bkoep

Thanks for the note... I can easily fold a surfing hot dog or helix bundle, tho I haven't master the perfect mutate yet. SPVincent shared a great barrel on the XL monomer -- quite nice

It would be great if you shared the most interesting shapes every so often. Compliments to players like SPVincent might encourage more of us to think a bit more creatively.

I'd recommend we get rid of the size penalty unless a smaller protein would be advantageous scientifically. Allow players to lop off a few segments, but most of us have realized that the maximum sizes score better. And let us know what you really want.... well bonded networks across pieces of the protein (like lots of sidechain bonds between helices), lots of h bonds in the center (I sort of figured that out on my own, tho I don't do them well). Something that will fold as a dimer or trimer? I sort of want to import some of my old folds to see if they will play well in a chain. Is that of value? Or is it something that will dock, or a barrel to penetrate a membrane or notch in a protein? Having a goal will help, and changing goals after 2-3 tries at it would keep this fun. the 2-3 tries give us time to refine our thinking

Just some crazy ideas...but no scientific background to know what is most useful


Joined: 05/19/2017
Groups: None
My suggestion was inspired by other randomizers!

If you folks are considering it, I figured I should give some background on it too! Some recent video games have had an increasing trend towards showcasing the randomizing of the game, whether it be on level design, difficulty conditions, key item locations, or acquisition of upgrades. 'Fortnite', 'Legend of Zelda' Randomizers, 'Dead Cells', and 'Into the Breach' are among the growing list of games featuring randomization as a core mechanic for difficulty/replayability.

A relevant example of this is the roguelike genre. In this genre players do "runs", or mini-campaigns. These typically involve dungeon crawling towards beating a series of bosses. No two playthroughs are ever the same however: All enemies, level rooms, upgrade items, and shop inventories are randomized, challenging the player to use whatever tricks at their disposal to fight their way through the unknown. They may see elements from one playthrough that were beneficial to them reappear in a subsequent run as a hindrance. Some games in the genre have even added additional challenge run conditions on top of the randomization difficulty, adding some options for the hardcore players to enjoy. 'The Binding of Isaac' even features a unique daily run mode complete with a leaderboard, as if the single player replayability wasn't enough already!

I've already gone over possibilities to take the inspiration from these other games and bring them into Foldit with the Objective Roulette idea, but I thought of one more way to shake things up (no pun intended). Instead of a straight chain with secondary structure Objectives, i.e. players may choose where they place their required SS's to fulfill the Objectives, the starting structure may be already locked into SS fragments at specific chain lengths. To put it differently, it is similar to the PSIPRED secondary structure hints from the unsolved de-novo puzzles, with the exception that they're unchangeable SS wise, but mutable sidechain wise.

Joined: 06/06/2013
Groups: Gargleblasters
randomized play in games isn't so new


believe it or not, there was a rescue the maiden sort of game I played while working at a division of Hewlett Packard in 1982. We had no visual, just text display and could move north, south, east, west, up and down only. There was a maze section which for each game had a random pattern set so each time you tried to get through was different :) The things design engineers did for fun :D


Joined: 05/19/2017
Groups: None
It's an old mechanic for sure, but there's new takes

Ah man, the good ol' days. I respect the text based adventures of old, those were an era I'm too young to fully appreciate, but I realize it's what made some of the things that I enjoy today what they are.


Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, RosettaCommons