Diffusion models, the same technology powering Stable Diffusion's image generation, are now being adapted to design entirely new proteins from functional specifications rather than just predicting how existing ones fold. Tools like RFdiffusion and ProteinMPNN generate protein backbones and amino acid sequences to order, producing candidate binders and enzymes that reach laboratory testing within weeks instead of the years that traditional physics-based design required.

What Are Diffusion Models and How Do They Work for Proteins?

Diffusion models learn to reverse a gradual noising process, a technique that has become foundational to generative AI. In image generation, these models start with random noise and iteratively refine it into a coherent picture. Applied to proteins, the same principle works in reverse: researchers begin with noise and denoise it step by step into a backbone with plausible bond geometry and a foldable shape.

This approach differs fundamentally from structure prediction tools like AlphaFold. A structure-prediction model takes an existing amino acid sequence and predicts the three-dimensional shape it adopts. A generative design model instead starts from a target function, shape, or binding site and produces a new backbone and sequence that could satisfy it. Most modern design pipelines link both approaches directly: a generative model proposes a backbone, a sequence design network assigns amino acids, and a structure predictor checks whether the resulting sequence is likely to fold as intended.

Which Tools Are Leading Protein Design Innovation?

RFdiffusion, developed by the Institute for Protein Design at the University of Washington, fine-tunes the RoseTTAFold structure-prediction network on a protein structure denoising task, transforming a model built to predict structure into one that generates it. The tool can produce protein backbones unconditionally or under specific constraints, including target binding sites, symmetric assemblies, and enzyme active sites.

In early demonstrations, RFdiffusion's developers reported designing protein binders against five separate targets, including influenza hemagglutinin, and used cryo-electron microscopy to confirm that one such binder's actual structure closely matched the computational design model. In a separate demonstration scaffolding the p53 helix that binds the MDM2 protein, the strongest designed binder bound roughly three orders of magnitude more tightly than the natural peptide it was built to outcompete.

ProteinMPNN, also from the Institute for Protein Design, is a graph-based neural network that predicts the amino acid sequence most likely to fold into a given protein backbone, one residue at a time, conditioned on its structural neighbors. It solves the critical step that RFdiffusion leaves open: a generated backbone has a shape but no sequence, and ProteinMPNN fills that gap. Its developers reported a sequence recovery rate of 52.4%, well above the 32.9% achieved by the physics-based Rosetta method it was benchmarked against.

Beyond these foundational tools, newer models extend generative protein design further. Chroma, developed by Generate Biomedicines, is a diffusion model that samples full protein complexes directly, using a graph neural network architecture built for more efficient computational scaling. Its developers reported experimentally testing 310 Chroma-designed proteins and confirmed two crystal structures with a backbone deviation of roughly 1.0 angstrom from the computational design, evidence that the sampled structures are physically realizable rather than merely plausible on paper.

How to Validate Protein Designs Before Laboratory Testing

• Self-Consistency Checks: Researchers predict the structure of a newly designed sequence with a separate tool and compare it back against the original design before committing resources to laboratory synthesis, ensuring computational plausibility.
• Experimental Confirmation: Validated designs undergo X-ray crystallography, cryo-electron microscopy, or direct biochemical assays to confirm that a protein actually folds and functions as intended in the laboratory.
• Sequence Recovery Benchmarking: Design pipelines measure how accurately sequence design networks recover natural amino acid patterns, with higher recovery rates indicating greater likelihood of successful folding.

A generated backbone is only a hypothesis, and researchers commonly perform self-consistency checks before moving forward. This validation step matters because a plausible in silico structure does not guarantee that a protein folds or functions as intended.

What Real-World Applications Are Emerging?

Generative protein design is moving beyond theoretical demonstrations into practical applications. RFdiffusion2, a follow-up model, extended active site scaffolding to work directly from atomic-level active site geometries rather than requiring residues to be pre-indexed along the backbone, solving all 41 cases in a diverse benchmark compared with 16 for the original method. The model produced working retroaldolase and hydrolase enzymes after screening fewer than 96 candidate sequences per reaction.

Current applications now include de novo binders, enzymes, and therapeutic candidates, with several designs validated by experimental methods. Extensions like LigandMPNN adapt the same architecture to design sequences around bound small molecules, metals, or nucleic acids rather than protein backbones alone, broadening the scope of what can be engineered.

A separate line of work applies diffusion directly to protein sequences rather than structures. EvoDiff, a framework from Microsoft Research, trains a discrete diffusion model on evolutionary-scale sequence data, aiming to generate viable protein sequences without requiring solved structures as a starting point.

The convergence of diffusion-based generative models with protein engineering represents a significant shift in how therapeutics and functional proteins are discovered. By borrowing the core mechanism that powers image generation tools, researchers are compressing timelines and expanding the design space of what proteins can be engineered to accomplish.