Diffusion probabilistic modelling of protein backbones in 3D | Jason Yim & Brian Trippe

If you enjoyed this talk, consider joining the Molecular Modeling and Drug Discovery (M2D2) talks live: https://m2d2.io/talks/m2d2/about/

Also consider joining the M2D2 Slack: https://m2d2group.slack.com/join/shar...

Title: Diffusion probabilistic modeling of protein backbones
in 3D for the motif-scaffolding problem

Abstract: The construction of a scaffold structure that supports a desired motif, conferring protein function, shows promise for the design of vaccines and enzymes. But a general solution to this motif-scaffolding problem remains open. Current machine-learning techniques for scaffold design are either limited to unrealistically small scaffolds (up to length 20) or struggle to produce multiple diverse scaffolds. We propose to learn a distribution over diverse and longer protein backbone structures via an E(3)-equivariant graph neural network. We develop SMCDiff to efficiently sample scaffolds from this distribution conditioned on a given motif; our algorithm is the first to theoretically guarantee conditional samples from a diffusion model in the large-compute limit. We evaluate our designed backbones by how well they align with AlphaFold2-predicted structures. We show that our method can (1) sample scaffolds up to 80 residues and (2) achieve structurally diverse scaffolds for a fixed motif.

Speakers: Brian Trippe -   / brianltrippe  

Jason Yim -   / json_yim  

Twitter Prudencio:   / tossouprudencio  
Twitter Therence:   / therence_mtl  
Twitter Cas:   / cas_wognum  
Twitter Valence Discovery:   / valence_ai  

~

Chapters:

00:00 - Intro
02:18 - Computational protein design workflow
10:57 - Diffusion models on protein backbones
13:13 - Forward diffusion and reverse denoising
20:32 - Why do diffusion models work?
21:29 - Why do diffusion for proteins?
23:59 - Model details
33:48 - Unconditional sampling
37:38 - Model limitations and failure modes
39:06 - Sampling SMCDiff
50:21 - Motif-scaffolding case studies and failure case
53:41 - Related work and conclusion
58:23 - Q+A