H-bond scoring?

Case number:699969-2001330
Topic:General
Opened by:LociOiling
Status:Open
Type:Question
Opened on:Monday, October 19, 2015 - 21:09
Last modified:Saturday, October 24, 2015 - 17:18

Some questions about how hydrogen bond networks are scored, especially in puzzle 1147b.

1. Does only one instance of a symmetric network score?
2. What does "over limit" mean?
3. How is the "polars satisfied" percentage calculated?

These questions come from looking at Beta Folders' best solution for 1147b, shown in "1147b symmetrical networks" (1147b.networks.png). Based on a retiredmichael solo, tt's very symmetrical, with the same network occurring twice.

As the combined image shows, the network on the right has 78.9474% of polars satisfied, and 9 interface h-bonds, with zero "over limit". It scores 710.

The network on the left has the same percent satisfied, but shows zero interface h-bonds, and says there are 7 "over limit". It scores nothing.

Both networks are exactly the same as far as I can tell. All the distances that can be measured between the bonded atoms are the same in both networks. Hence question 1, why does only one instance of the network have a score?

The second question follows directly from the first. Both instances of the network are the same, but one is zero over limit and the other is seven over limit. What does "over limit" mean here, and how is it calculated?

The third question is about the "polars satisfied" percentage. The "1147b network detail" (1147b.detail.png)shows a close-up of the network that scored. Except for segment 9, atom 8, all the bondable atoms have at least one bond.

The wikia page http://foldit.wikia.com/wiki/Hydrogen_Bond_Networks has some helpful suggestions about donors and acceptors, but in this example, I'm not finding a numerator and a denominator that get me to 78.9474%.

So, third question, is it possible to have a short summary of how "polars satisfied" is calculated?

Some follow-ons to the third question:

Is scoring done atom-by-atom, or does it look only at the number of bonds per residue?

Can the potential score for each residue be listed in a simple static table, or is it dependent on what the residue is bonded to?

Can the segment info be updated include the bonding percentage? (Oops, stealth enhancement request.)

Any light that could be shed on these perplexing questions would be greatly appreciated.

AttachmentSize
1147b symmetrical networks660.25 KB
1147b network detail677.45 KB
(Mon, 10/19/2015 - 21:09  |  11 comments)


jflat06's picture
User offline. Last seen 2 days 19 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

The 'limit' is an upcoming feature for HBond networks - it isn't actually affecting your score in this puzzle.

The reason that the network on the left isn't getting any score is because it is a copy of the other network. Previously, we showed the copy networks as having some score. This was incorrect, however, as you only ever got points for the original network. With the new setup, your total points for the filter will be exactly the sum of each of the networks. This will be important for the upcoming limit feature.

bkoep's picture
User offline. Last seen 2 days 6 hours ago. Offline
Joined: 11/15/2012
Groups: None

Perplexing questions, indeed!

To add to jflat06's answer, the "over limit" feature was recently added so that we can discourage massive, high-scoring networks at the interface. Large numbers of interface H-bonds are actually detrimental to binding, so we're trying to encourage small, high-quality (i.e. completely satisfied) networks.

When evaluating the satisfaction of a network, every atom capable of forming an H-bond is considered—even individual hydrogen atoms! I think the case shared above will make more sense if you use the "Stick+polarH" view option (accessible if you've checked Menu->General Options->Show advanced GUI). Lysine, for example, has only one nitrogen atom shown, but that nitrogen has three hidden hydrogen atoms, each of which need to be satisfied.

Because the score of a network is not a simple "sum of parts," you wouldn't be able to break it down into per-residue components like you suggest. However, it could be possible to include each residue's "H-bond satisfaction" in the segment info; would that be helpful?

Also, while the "over limit" feature did not apply in Puzzle 1147b, it is currently active in Puzzle 1150!

Susume's picture
User offline. Last seen 17 min 41 sec ago. Offline
Joined: 10/02/2011

Stick+polarH view shows how many bonds a donor sidechain can make (how many white H atoms it has), but not how many bonds an acceptor can take. And some (serine, threonine) you can see the white H and think it can only make one bond, but it is also an acceptor and can make more.

LociOiling's picture
User offline. Last seen 1 hour 5 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders

Thanks for the tip, bkoep, stick + polar H did the trick, getting me to the magic 78.9474% shown in my example.

The stick + polar H view takes a bit of getting used to. If combined with EnzDes or CPK coloring, the hydrogens show up as white tips. You'll also see white hydrogens on the backbone, and it can be hard to distinguish the sidechains at first. If you're looking at a helix, turning on "Show bonds (helix)" makes it easier to see which hydrogens are in the backbone. Then turning "Show" for the HBond Network filter helps distinguish which hydrogens are relevant to the network.

Sidechains like aspartate and glutamate won't have any white hydrogen tips, but arginine and lysine will.

Back on the "over limit" feature, could we get a clear statement on what's counted toward the limit? Is is the number of bonds? The puzzle 1150 comments are kind of confusing on this topic.

bkoep's picture
User offline. Last seen 2 days 6 hours ago. Offline
Joined: 11/15/2012
Groups: None

Correct, as Susume mentioned above, hydrogen bond acceptors (like the oxygen atoms of aspartate or glutamate) do not indicate the number of H-bonds they can accept. For simplicity, Foldit considers acceptors to be satisfied if they make any hydrogen bonds.

In Puzzle 1150, the filter will stop rewarding networks after it has counted three inter-molecular H-bonds (totaled among all networks). If you have built several networks, the filter will start scoring the most satisfied networks, accumulating inter-molecular H-bonds and stopping once it reaches the limit. In the end, we want players to insert only a handful of hydrogen bonds across the interface, and then to build networks around those inter-molecular bonds such that all polar atoms are satisfied.

Again, I understand that the mechanics of the HBNet Filter are opaque and difficult to decrypt—I'm sorry that we haven't provided more thorough explanations. However, your questions are really helpful to us in pinpointing the concepts that need to be clarified! I'm hoping to put together an explanatory video and blogpost soon!

frood66's picture
User offline. Last seen 5 hours 31 min ago. Offline
Joined: 09/20/2011
Groups: Marvin's bunch

an H bond satisfaction tool would be great. - just click a button to show those not satisfied.

Joined: 10/30/2012
Groups: Beta Folders

not sure how helpful the segment info would be,
might be worth a try for a couple of puzzles.
we could verify what we have done and experiment with combinations.
then build something around those combinations

big maybe

Susume's picture
User offline. Last seen 17 min 42 sec ago. Offline
Joined: 10/02/2011

I would like to see the % satisfied for each segment in the Tab window. Others would prefer glowy things on the unsatisfied sidechains when "show" is checked on the filter, for easy visibility.

Joined: 06/24/2008
Groups: Void Crushers

I find it impossible to see stick + polar H; there is too much.

Joined: 06/24/2008
Groups: Void Crushers

I find it impossible to see stick + polar H; there is too much.

spvincent's picture
User offline. Last seen 1 hour 2 min ago. Offline
Joined: 12/07/2007
Groups: Contenders

Another H-network scoring quirk: a bond that supposedly doesn't score well enough to be part of the network (displayed as a red barber pole) nevertheless contributes to the percentage polar satisfied score.

See scientist shared solution "spvincent scoring quirk". Mutating Arg 39 to something else reduces the score of the associated network to below the 75% threshold, even though the only contribution made by that Arg sidechain to the network is a single bad-scoring (i.e. red) H bond.

Sitemap

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, Microsoft, Adobe, RosettaCommons