The fires model (ProteinMPNN) appears to be more or less the reverse of 
AlphaFold, as John says. Given the 3D structure of a protein, it attempts to 
find a sequence of amino acids that would fold into that shape. Algorithms 
already existed in this domain, but they managed to improve accuracy using deep 
learning (from 32.9% to 52.4%). I have no idea how "good enough" the latter 
number is, and if there is more room for improvement with deep learning alone.

Then the second paper is about a generative model, that provides a protein 
given some high-level features (number of protomers and protomer length). My 
understanding is that this allows one to design a "key" that fits some specific 
"lock". Generative models are capable of learning the mapping from one feature 
space to another (for example, from images of apples to textual descriptions of 
apples  e.g. "big red apple") and then operate in reverse. Here they must have 
trained the model with known 3D structures and the associated features they are 
interested in, and can now "hallucinate" new molecules from sets of features 
that the model has never seen before, but that allow for some novel application.

Telmo

Am Fr, 16. Sep 2022, um 19:03, schrieb John Clark:
> AlphaFold solved the protein folding problem some time ago and now, judging 
> from two articles in today's issue of the journal Science, it looks like the 
> inverse problem has also been solved, the protein design problem. If you tell 
> a program called "ProteinMPNN" that you want a 3-D protein that has an 
> activation site that will perform a very specific function, and has the 
> proper scaffolding to keep that activation site stable, and is also shaped in 
> just the right way so that it can fit into a very tight corner where it is 
> needed like a key into a lock, then ProteinMPNN will tell you what linear 
> sequence of amino acids will fold up into that 3-D shape. I think this is a 
> very big deal, the implications for medicine are obvious but it also 
> signifies a huge advance in Nanotechnology because the authors claim the 3-D 
> shape the sequence of amino acids folds up into is within 0.06 Nanometers of 
> the requested shape, and Nanotechnology is about placing atoms exactly where 
> you want them to go, and enzymes are proteins and they act like little 
> machines.  
> 
> Robust deep learning–based protein sequence design using ProteinMPNN 
> <https://www.science.org/doi/10.1126/science.add2187>
> 
> Hallucinating symmetric protein assemblies 
> <https://www.science.org/doi/10.1126/science.add1964>
> 
> John K Clark
> 
> 
> 
> 
> 
> 
> -- 
> You received this message because you are subscribed to the Google Groups 
> "Everything List" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to everything-list+unsubscr...@googlegroups.com.
> To view this discussion on the web visit 
> https://groups.google.com/d/msgid/everything-list/CAJPayv2kQNjwf6pB-uxG1TDoGbMy8orDMOXKq%3DiXyzEhB%3DADXQ%40mail.gmail.com
>  
> <https://groups.google.com/d/msgid/everything-list/CAJPayv2kQNjwf6pB-uxG1TDoGbMy8orDMOXKq%3DiXyzEhB%3DADXQ%40mail.gmail.com?utm_medium=email&utm_source=footer>.

-- 
You received this message because you are subscribed to the Google Groups 
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to everything-list+unsubscr...@googlegroups.com.
To view this discussion on the web visit 
https://groups.google.com/d/msgid/everything-list/b756b4a5-b31a-4340-b749-6028432b49cf%40www.fastmail.com.

Reply via email to