Diffusion Probabilistic Modeling of Protein Backbones in 3D for Motif-Scaffolding

Explore diffusion probabilistic modeling of protein backbones in 3D for motif-scaffolding in this comprehensive talk. Learn about computational protein design workflows, diffusion models applied to protein structures, and the SMCDiff algorithm for efficient scaffold sampling. Discover how this approach can generate diverse scaffolds up to 80 residues long while supporting given motifs. Examine model details, unconditional sampling, limitations, and case studies. Gain insights into the potential applications for vaccine and enzyme design through this cutting-edge machine learning technique presented by experts Jason Yim and Brian Trippe.

- Intro
- Computational protein design workflow
- Diffusion models on protein backbones
- Forward diffusion and reverse denoising
- Why do diffusion models work?
- Why do diffusion for proteins?
- Model details
- Unconditional sampling
- Model limitations and failure modes
- Sampling SMCDiff
- Motif-scaffolding case studies and failure case
- Related work and conclusion
- Q+A