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

By Brian L. Trippe, Jason Yim, D. Tischer, Tamara Broderick, David Baker, Regina Barzilay, Tommi S. Jaakkola. 2022

347 citations recorded by the source corpus.

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.

Subfields: Information Systems, Molecular Biology, Generative Models, Computational Biology & Bioinformatics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Information Systems

In Molecular Biology

In Generative Models

In Computational Biology & Bioinformatics

Citations in the corpus

Generalized Biomolecular Modeling and Design with RoseTTAFold All-Atom (2023, methodology): “The standard method for in silico assessment of protein generative modeling is to measure the RMSD between the generated protein backbone and AF2’s prediction of that backbone from sequence alone (42)-(46).”
Hallucinating symmetric protein assemblies (2022, methodology): “Computational requirements and hardware memory limitations become bottlenecks for hallucination of increasingly large structures; the development of computationally less expensive structure prediction methods with fewer parameters, as well as generative approaches such as diffusion models (37, 38) t”
Equivariant 3D-conditional diffusion model for molecular linker design (2022, neutral): “More details on datasets, baselines and metrics can be found in Methods.”
Diffusion Models: A Comprehensive Survey of Methods and Applications (2022, background): “[267] proposes to first learn a distribution over diverse and longer protein backbone structures via an E(3)-equivariant graph neural network, and then efficiently samples scaffolds from this distribution given a motif.”
Broadly applicable and accurate protein design by integrating structure prediction networks and diffusion generative models (2022, result): “This definition of success is significantly more stringent than those described elsewhere (refs [5,8,16,25], Fig.”
Illuminating protein space with a programmable generative model (2022, neutral): “24,25), or cubically, O ( N 3 ) (refs.”
DiffusionDet: Diffusion Model for Object Detection (2022, background): “…models have recently demonstrated remarkable results in fields including computer vision [4,21,33,35,39,68,71,75,76,82,104,108], nature language processing [3,30,52], audio processing [41,48,50,70, 91,100,103], graph-related topics [42], interdisciplinary applications [2,40,44,78,94,99,102], etc.”
Protein structure generation via folding diffusion (2022, methodology): “Most importantly, since each residue forms its own independent reference frame, there is no need for an equivariant neural network, as required by diffusion models that work on Cartesian coordinates of protein structure 14,15 .”
A Survey on Generative Diffusion Models (2022, neutral): “Folding DiffFolding[248]/Trippe et al.[249]”
SE(3) diffusion model with application to protein backbone generation (2023, methodology): “One important research direction is to extend FrameDiff to conditional generative modeling tasks, such as probabilistic sequence-to-structure prediction which is key to capture functional motion (Lane, 2023), but also probabilistic scaffolding design given a functional motif (Trippe et al., 2022) In”
Practical and Asymptotically Exact Conditional Sampling in Diffusion Models (2023, positive): “Evaluation: We use self-consistency evaluation approach for generated backbones [29] that (i) uses fixed backbone sequence design (inverse folding) to generate a putative amino acid sequence to encode the backbone, (ii) forward folds sequences to obtain backbone structure predictions, and (iii) judg”
Diffusion models in medical imaging: A comprehensive survey (2022, methodology): “The platforms presented in [74, 139] can be used for designing drugs and vaccines.”