AlphaFold Mutate tool

Case number:699969-2011976
Opened by:nspc
Opened on:Thursday, August 19, 2021 - 19:13
Last modified:Monday, August 23, 2021 - 19:11

Is it possible or is it planned to use the AlphaFold technology to do the opposite?
I mean to find a AA sequence from a protein folding instead of find a protein folding from a AA sequence.

It would be a like an "AlphaFold Mutate"

It can be usefull to help us to find better Amino acids that have more chance to keep the folding in lab.

(Thu, 08/19/2021 - 19:13  |  2 comments)

Joined: 10/10/2015
Groups: Team China

Try my neuroNet recipe for now

bkoep's picture
User offline. Last seen 7 hours 4 min ago. Offline
Joined: 11/15/2012
Groups: Foldit Staff

Very interesting suggestion! Yes, in theory, I believe there could be ways to use the AlphaFold neural network to suggest mutations for a provided backbone. I am thinking of something similar to the "gradient backpropagation" method that was used with the trRosetta neural net (described in this paper).

Unfortunately I do not have the ML expertise to say more, and I am not sure if the AlphaFold network would need to be retrained to support this, or how the network's resource requirements might change. But other people are looking into this, so stay tuned!


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