Oke, the AlphaFlow paper is awesome: AlphaFold Meets Flow Matching for Generating Protein Ensembles
Just watch how AlphaFlow's ensemble reproduces details of MD.
Weights + code
We have it in the reading group on Mon 11am EST!
1/2
@HannesStaerk
AlphaFlow and Distributional Graphormer are very similar for proteins. Flow matching vs diffusion, slightly different training data, but essentially the same representation and architecture. AlphaFlow is better validated and presented indeed.
@FrankNoeBerlin
We also had Distributional Graphormer in the reading group! I suppose an important part of AlphaFlow is also training on the ATLAS dataset which was only released in Nov 2023 after Graphormer was published
@HannesStaerk
Thanks for starting this conversation. I've added the paper to my to-read list.
I've been evaluating these methods as I would a new enhanced sampling algorithm. Big questions are... 1/n
@HannesStaerk
It is super nice to be able to generate fluctuations similar to those in equilibrium MD simulations. I am curious whether it can predict large-scale functional conformational changes, like those demonstrated recently on KaiB by AlphaFold-2 + MSA clustering.
@HannesStaerk
Can this be done with AlphaFold-Multimer to better understand protein-protein interactions? Could be a really great next step and there are so few models that give good insight into PPIs and binding affinities. Perhaps there is some way to use ipTM scores for ensembles here?