Protein structure generation via folding diffusion

TL;DR

This paper introduces a diffusion-based generative model for protein backbone structures that mimics natural folding, producing realistic, novel proteins efficiently without complex equivariant networks.

Contribution

The authors develop a new diffusion model that generates protein structures by simulating the folding process, simplifying the architecture and improving realism.

Findings

Unconditionally generates realistic protein structures.

Uses a folding-mimicking diffusion process.

Achieves structural complexity comparable to natural proteins.

Abstract

The ability to computationally generate novel yet physically foldable protein structures could lead to new biological discoveries and new treatments targeting yet incurable diseases. Despite recent advances in protein structure prediction, directly generating diverse, novel protein structures from neural networks remains difficult. In this work, we present a new diffusion-based generative model that designs protein backbone structures via a procedure that mirrors the native folding process. We describe protein backbone structure as a series of consecutive angles capturing the relative orientation of the constituent amino acid residues, and generate new structures by denoising from a random, unfolded state towards a stable folded structure. Not only does this mirror how proteins biologically twist into energetically favorable conformations, the inherent shift and rotational invariance of this representation crucially alleviates the need for complex equivariant networks. We train a denoising diffusion probabilistic model with a simple transformer backbone and demonstrate that our resulting model unconditionally generates highly realistic protein structures with complexity and structural patterns akin to those of naturally-occurring proteins. As a useful resource, we release the first open-source codebase and trained models for protein structure diffusion.