Request: Sequence Suggestion Tool (reverse-remix)

Case number:845813-997106
Topic:Game: Tools
Opened by:brow42
Opened on:Monday, February 24, 2014 - 21:03
Last modified:Tuesday, February 25, 2014 - 22:26

Design is hard. We can make new structures that score well in rosetta, but are not stable in the lab. The remix tool was presented as a way to make more stable structures. The remix tool, as I understand it, takes our selected AA sequence, checks for any occurrences of that sequence in the fragment library, and reshapes our selection to match the known fragment.

Hopefully the new shape is close to our old shape. Often it's not. Remix almost always has to be used with the cut tool. Healing the cut often returns us to a shape that remix wants to change, again. Often it fails, because our sequence isn't in the library.

The remix tool changes the shape but preserves the sequence. I think players are much more attached to the shape they are designing and don't care at all about the sequence. I think players mostly only care whether residues are hydrophobic or hydrophillic, and only to the point that they form the core, the surface, or an interaction area. The remix tool is precisely the opposite of what is needed here.

The mutate tool should do this, but it does not work well. By well, I mean that does not find mutations that a brute-force tool will find (with huge gains!). The mutate tool stops finding mutations very quickly.

The proposed tool should work as follows:

1) The 3-D structure of the selected region should be converted into a quantized canonical form. There are many examples in the literature, however, they are not all appropriate for this purpose (large window length, optimized symbol entropy, etc.). You may be able to do this with your current fragment library on the fly, or you may have to do a fragmet->structure conversion and save it as a resource.
2) A (fuzzy) search for this structure in the structure library. Fuzzy because similar shapes are acceptable.
3) Return an AA sequence that is similar to what the player already has. At this point, the only criterion I can think of is preserving 1-to-1 hydrophobic classification, or the overall "hydrophobic moment" of the selection.

I think this tool will allow players to be creative while still improving the chances of our designs being stable.

(Mon, 02/24/2014 - 21:03  |  1 comment)

bkoep's picture
User offline. Last seen 4 days 7 hours ago. Offline
Joined: 11/15/2012
Groups: None

I'm a big fan of this idea. We do this kind of thing frequently in the Baker Lab on our own design projects, and I do think it could improve the success rate for Foldit designs on the bench.

To clarify, the current Remix tool does not take into account sequence information when searching for a replacement backbone—it only considers the endpoint coordinates of the selection. As a result, you often get a replacement fragment with little concordance between the backbone shape and the amino acid sequence. We hoped the mutate tool would resolve the worst of the problems, but the most intuitive solution would indeed be to just grab the native sequence from the replacement fragment in the structure library.

The Remix tool has paved the way for coding a reverse-remix into Foldit, but this has been put on the back burner for the time being. I hope we'll be able to reopen this issue soon!


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