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

TL;DR

This paper introduces SMCDiff, a diffusion-based graph neural network model that generates diverse, longer protein backbone scaffolds conditioned on motifs, advancing scaffold design for vaccines and enzymes.

Contribution

It presents a novel E(3)-equivariant diffusion model capable of sampling diverse, longer protein scaffolds conditioned on motifs with theoretical guarantees.

Findings

Samples scaffolds up to 80 residues long

Achieves structural diversity for fixed motifs

Aligns well with AlphaFold2-predicted structures

Abstract

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.