5 replies [Last post]
Joined: 03/28/2020
Groups: Go Science

I would like to open discussion on what a scientifically valuable solution in symmetric puzzles actually is.
The reason for this is that even though I have played many of them now I do not really know what actually is a scientifically "good" solution.

If it would just be for the score then I would only do helix or sheet+helix bundles that are packed as much as possible to each other. For example in 1977 Tetramer the to me seemingly straight-forward approach is to design a "surfing hotdog" (sheet-region on top of two helix bundles). This will give a somewhat "brick-like" quadar which can be positioned so that it packs together nicely with the symmetric chains (so that basically no voids are present). To satisfy the HBond-objective, the simplest way is to put in 3 histidines close to the symmetric origin. The histidines will EXACTLY satisfy the HBond-objective once they bond properly to the symmetric copies.

Even though this seems to be the simplest way to solve these puzzles, I get the impression that it is not the scientifically most valuable and I would like to understand why? It is to me as regular as this symmetric structure can be. Of course there are large contact-areas that may possible not exactly fold up as modeled in Foldit but to me this can also happen with other approaches.

Of course many other approaches can be taken and many Foldit players have come up with really great solutions. Let it be skewed bundles or barrel-like connected sheets. The newsletters gave a good impression of them.

But what has real worth for science? Is it as much as possible buried HBond-networks with as few as possible BUNs? Or as LITTLE as possible contact area between the main monomer and symmetric copies? Or should the bundles be skewed as much as possible? Should it be as regular/packed as possible?

I would really be thankful for some guidance there. I know that much is possible but to me it would be more effective to better understand in what direction we should be looking.

A further thing is the HBond-bonus-function. I have been playing around a lot with HBond-networks as of late and I do not understand fully how the HBond-bonus is actually derived. Sometimes I got nets that are huge but the bonus was not full even though there were many connections between symmeric chains and main and few open atoms/BUNs. I find it hard to EXACTLY satisfy the HBond-bonus limit without doing the 3xhistidine in the center approach. Even though in most cases it is possible in skewed designs after some fiddling with the net and in most cases DELETING parts of the net. Is this really a desired behavior for the HBond-bonus function?

Thanks in advance and happy folding to all!

Joined: 06/06/2013
Groups: Gargleblasters
Additional questions and subjects

I have had less scoring success on these puzzles, in part as I do not do well hand mutating. Particularly with my mouse had out of commission after shoulder surgery. I should note that I am a "lego block" sort of player without a strong science background. That noted, I have a personal bias towards the following objectives:

1)Try not to match the same structure with itself, e.g. a helix AAs 1-14 with helix AAs 1-14. This is not always possible to avoid due to the type of symmetry. It can be done more easily with helix bundles than surfing hot dogs
2)I strive to get my h bond networks close to the core of the protein, and if possible get at least 2 network planes
I avoid the threonine, serine, histadine self connecting networks as they strike me as unlikely. Threonine and serine in particular are likely to bind to anything, so I don't think they will necessarily work as a unique binding network that will make the symmetric parts link
3)I have decided that after a successful computer designed anti-inflammatory binder had some BUNS a couple months ago that I would not reject a design for a small number of buns
4)If a design is unstable, e.g. loops go wonky easily, it is not a good solution

That said, I'd love to know how other design players think about a "good" solution. I am not sure that the scientists know at this point, so perhaps it is better to not have us all try to think about these puzzles the same way. Variety may yield more than one type of design compromise that works. Sometimes the mojo just feels right, even if the score is not the best. If the mojo is good, I pursue it anyways even if it will not score competitively. I would actually find it useful to work the same puzzle for longer and really pursue alternatives or try a shape repeatedly to see if I can make it work

As for h bond points, they count only if they cross monomers at least once, and at least 75% (or 66%, I forget which) of the hydrogens have been paired up with a sidechain bond. Tyr, Threonine and Serine will mate with anything and have only one hydrogen. Hence they form triangles or little circles that are 100% satisfied which makes for more points than a network that has one unsatified hydrogen hanging off. Histadine can also form a neat circle if both hydrogens are paired. JMBrownlee or Susume might help us understand if these bonds are likely to occur in nature, as would any other biochemistry pro

Good questions. It will be interesting to see what other answers or heuristics some of our creative folders use


Joined: 03/28/2020
Groups: Go Science
You brought on some very good

You brought on some very good points there!
On(4): I fully agree with this. If a design is unstable in itself (in that it produces many BUNs when wiggled or loops on the edge of ideality) then this design is probably not good. I tried many things in this regard and for me the outcome always was that in the initial design-phase you should take as much care as possible so that no part of the protein "twists" too much. If you achieve that then you bascially always get a very high-scoring result in the end (if you do not screw up anything on the way to it :o) ). A side-effect is that optimization-recipes don't like unstable designs at all because the objectives are toggling and that extends dramatically the time to get an optimized solution or makes overall optimization impossible. A rule-of-thumb for me is that after I have basically finished my design (so folded and placed to target and some semi-manual optimization) and if I run cut&wiggle then and it gains good points up to about 30mins, then the design is pretty good. Because with c&w the objectives will easily toggle quite a bit.
On(3): Agreed. Even though I favor myself fully satisfied nets but sometimes the unsatisfied ones look nicer and can often be larger and have more 3-bonders in. Sometimes the BUNs in it are anyway just on the verge of being a BUN and so this may not be so problematic.
On(2): There I am not sure. From what I have seen, HBnets which are very much buried do often not score as well as ones that start in the interface and "grow" outside of it. These will still give good score but will be not that much in the iterface. I think this is because we are hiding hydrophylics there which actually don't really like to be hidden even though they may be perfectly bonded. I sometimes had setups where the deletion of the net or large parts of it lead (after optimization with recipes) to an actually higher score.
Yes, especially serine is pretty "flexible" in regard where it will bond to. Threonine is OK for me and it always gives good bonus once it has been well-bonded. I have become to love Tryptophan because it can perfectly terminate a branch in the net with an open "red". Serine or likewise are often "too short".
That is actually one of my base-questions if 3x histidine at the symmetric center is a viable solution. I had the best scoring results with that but will it actually fold up this way in reality? Since the (hydrophobic) interface is very large then there may be a high chance that things will "push each other away" if they are not exactly aligned as in simulation. So is this approach really something to pursue in the future? For score I am certain it is.

Variety and diversity is always a good thing and that was communicated well up to now. Especially with the last binder-challenge I think this topic came out very clear. I sometimes also thought that a longer time to try out things for a single puzzle might be good but I am also always looking forward to new puzzles because of a new challenge. I think overall I am happy with the frequency of puzzle releases.

With Hbond-networks I am still a bit puzzled sometimes but the overall goal should really be to satisfy them as much as possible. This will almost always get better score. With the 3-bonders like Glutamine this is often not a simple matter but I understand that they will be a much better unique "key in the lock" than a histidine-circle or so because it simply only fits on way into the net.

Thanks Skip for bringing these things up!

bkoep's picture
User offline. Last seen 9 hours 56 min ago. Offline
Joined: 11/15/2012
Groups: Foldit Staff
Guidance for symmetric design

First, I want to be clear: It is always good to play for score.

This is a central tenet of Foldit. You don't need to understand any of the science to contribute, because the Foldit score itself should tell you whether your solution is getting better or worse. If the Foldit team sees that the highest-scoring solutions are worse than lower-scoring ones, then it's up to the Foldit team to adjust the score function.

It's true that the score function isn't perfect. We sometimes talk about short-comings of the score function, and we recently ran a whole experiment on the premise that limitations of the score function can lead to inefficiency in design. But the Foldit project is designed so that you can play for high scores contributes to protein research. If you have the impression that playing for score is not a good way to help Foldit research, then that sounds like a communication failure on our part.

With that out of the way, we can get to the real issue, which seems to be about guidance in symmetric design puzzles.

As great as the Foldit score function is, it does not offer much "big picture" guidance, or suggest what you might aim for in a puzzle. Unfortunately, I won't tell you precisely what to aim for, because (as Skippysk8s noted) I myself don't know the answer, and we do like the diversity that comes from lots of players with different ideas.

However, maybe it will help to share my own perspective. I'd like to focus on this part of ichwilldiesennamen's post.

But what has real worth for science? Is it as much as possible buried HBond-networks with as few as possible BUNs? Or as LITTLE as possible contact area between the main monomer and symmetric copies? Or should the bundles be skewed as much as possible? Should it be as regular/packed as possible?

In so many words, you are circling a major theme in symmetric protein design. And perhaps it will be helpful to focus on this challenge when you are designing symmetric proteins: reaching the correct balance of polar and hydrophobic residues on the surface of the monomer unit.

The experimental evidence is pretty clear. With too many hydrophobics on the surface, the monomer unit is likely to misfold and aggregate with other off-target material. With too few hydrophobics on the surface, the monomers may fold, but they will not stick to one another to form a symmetric assembly.

If we omit the Objectives, the base Foldit score will encourage large, completely hydrophobic interfaces, which puts lots of hydrophobic residues on the monomer surface. We've tried a couple different strategies to counter this tendency and reduce the hydrophobic surface of the monomer unit, such as: reducing the size of the interface, or placing (satisfied) polar residues at the interface to form a network of H-bonds.

(You could also imagine a simple Objective that directly limits the proportion of polar/hydrophobic residues in a design. Paradoxically, this "simple" solution has complex ramifications for the Shake and Mutate algorithms. However, we may have an update along these lines in the future...)

I don't know if that will provide any guidance on the symmetric design puzzles, but I would encourage you to think about different ways to balance the amount of hydrophobics on the monomer surface. By and large, the top-scoring designs from recent symmetry puzzles do seem to look pretty good in this respect. So you may rest assured that playing "for score" still produces great solutions in our symmetric design puzzles.

Joined: 03/28/2020
Groups: Go Science
This is very helpful!

Thanks bkoep for this information!

I know that it was often stated that the score-function of Foldit is quite reliable and we can and should trust it. And quite frankly, it is for me quite fun to just play for score and not to worry too much about the biochemistry background behind it. This was one of the main reasons to get me going with Foldit in the first place because I can simply consider each puzzle as an optimization-challenge. But at some point it would be good to know if the direction one has taken in the past is really also the way to go for further designs. And this is basically my question with the 3x histidine-approach at the symmetric center.
But since you write that score is (or can be) most relevant I will persue this in the future (as well) because it really is well-scoring. So if this may not be wanted anymore in the future I will wait for Foldit-staff to forbid it :)
I was rather wondering if doing always the same approach would still be helpful for science. But there also each design is not exactly the same and maybe this minute diversity may also be helpful I guess.

I somehow feared the answer that it is not exactly known what is actually helpful or not. But I guess that is to some extent the nature of science. But your emphasis on the right balance hydrophob<=>hydrophyl is helpful. I remember that it was mentioned some times but it probably blurred out of my mind the last days :o) .
So this basically means that it is probably not THAT useful to have extremely large HBnets in the interface. Had been experimenting with this and in most cases it is even possible to substitute ALL hydrophobs in the interface to hydrophylic but it is clear to me as well that then the interface is basically the same as the surrounding water so that there would actully be no need anymore to bind to the interface instead of the water.
So the underlying request is understood but the next question directly results out of this: what is considered a good balance (in percent)? With the 3x histidine there is probably 80% hydrophob at the interface. With larger buried nets this may be only 50% or less. Is there any thumb-value for this? THAT might be helpful.
A simple way to limit the interface area overall is to skew the monomer against the symmetric copies. So I understand that this is probably also a desired way to go.

Until then I will keep on focussing on score and try a couple of new things while keeping the interface-balance in view. Thanks for outlining this point so well and bringing it back more to mind again.


bkoep's picture
User offline. Last seen 9 hours 56 min ago. Offline
Joined: 11/15/2012
Groups: Foldit Staff
H-bond Networks

The H-bond Network Objective is tricky, and can behave counterintuitively if you have low-scoring (red) bonds in your network, or if the network is less than 100% satisfied, or if residues in your network are not in the protein core. There are some more details about the H-bond Network Objective in this blogpost.


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, Boehringer Ingelheim, RosettaCommons