Back to paper
Highly accurate protein structure prediction with AlphaFold
Question

Which part of the model most directly uses evolutionary information?

DOdortiz· 10 days ago

As someone from outside structural biology: the MSA (multiple sequence alignment) is where co-evolution signal lives, right? Is it the Evoformer's MSA-row/column attention that turns 'these residues co-vary across species' into 'these residues are in contact'? Trying to build the right mental model.

Bioinformatics

1 Reply

Sign in to reply and react.

Accepted answer

HChcole10 days ago

Exactly — co-evolving residues are often in physical contact, and the MSA representation + pair representation let the network reason over that. The structure module then turns the learned pair geometry into actual 3D coordinates.