SE (3) diffusion model with application to protein backbone generation

Reciept of FrameDiff

1. Train two flow diffusion model: Riemannian diffusion for residue rotation matrices and Gaussian diffusion for residue translations.
2. Use best practices from AlphaFold2.

Methodology

The authors use the same parameterization as in AlphaFold2 and RFDiffusion. The protein is represented as a sequence of residue rotation and translation.

The diffusion process is essentially the same as in RFDiffusion. The main difference is the absence of folding pretraining.

I recommend the presentation of Jason Yim, which compares FrameDiff, RFDiffusion, and FrameFlow.

@article{yim2023se,
  title={SE (3) diffusion model with application to protein backbone generation},
  author={Yim, Jason and Trippe, Brian L and De Bortoli, Valentin and Mathieu, Emile and Doucet, Arnaud and Barzilay, Regina and Jaakkola, Tommi},
  journal={arXiv preprint arXiv:2302.02277},
  year={2023}
}