Make blueprint mutate protein so it will Alphafold out of the box

Case number:845813-2013116
Topic:Game: Tools
Opened by:mwm64
Status:Open
Type:Suggestion
Opened on:Sunday, May 1, 2022 - 19:08
Last modified:Monday, May 2, 2022 - 18:23

Right now if you use the blueprint tool, it shapes your protein, but it does not
mutate it appropriately -- in particular loops are generated with all "i" residues,
which AlphaFold completely ignores. If the blueprint tool mutated the loops be Gly etc.
when it made them so they would actually be a) ideal loops and b) ones that AlphaFold
would recognize, as a loop transition, it would make life much easier. Similarly if
sheets are mutated to a segment type that AlphaFold knows are sheets and not helices,
that would be nice.

Basically, the test is: if I make a 4 by 4 sheet with two helixes on the back in
the blueprint tool, and submit it to alphaflow, it should really fold up that way.
As its stands, no matter what you make in the blueprint tool, it comes out as one
long helix in alphaflow.

(Sun, 05/01/2022 - 19:08  |  1 comment)


bkoep's picture
User offline. Last seen 5 hours 31 min ago. Offline
Joined: 11/15/2012
Groups: Foldit Staff

If only we had such a tool!

Unfortunately, this technology does not exist yet. There is not a segment type that unambiguously encodes for "this segment should be a sheet". Even if you know which residues are most common for a particular loop shape, there is no guarantee that those residues are compatible with your overall fold. The actual energy landscape of your protein (and the AlphaFold prediction!) depends on the entire protein sequence. We do not have an algorithm that can back-calculate a design sequence that AlphaFold will like.

That said, there have recently been some major advances in using neural networks for protein design. We are currently working on some updates to the Mutate tool that may get us closer to what you are asking for.

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, Amazon, Microsoft, Adobe, Boehringer Ingelheim, RosettaCommons