Duke University School of Medicine researchers have unveiled Raygun, an artificial intelligence framework that can redesign proteins at a scale reminiscent of natural evolution. Unlike conventional protein engineering, which often starts from scratch or makes limited changes, Raygun can generate shorter, longer, and extensively modified protein variants while preserving key structural and functional features.

Raygun is built on protein language models—AI systems trained on millions of protein sequences. These models learn statistical patterns that link amino-acid sequences to three-dimensional structure and biological activity, acting as a kind of “translator” between sequence information and function.

What makes Raygun distinctive is its new way of representing proteins. Instead of treating sequences purely as variable-length chains, the method converts proteins into a standardized mathematical format that captures patterns learned by the AI. This unified representation allows the system to compare proteins across sizes and to redesign them without losing the information required to maintain their core behavior.

Researchers can control Raygun with two settings: one regulates how strongly the output sequence diverges from the original, and the other determines whether the protein becomes shorter or longer. By tuning these parameters, the tool effectively supports targeted “size programming” alongside purposeful sequence modification.

In validation experiments, Raygun-generated proteins showed predicted structural integrity and retained important functional sites. The team then tested performance in living cells using fluorescent proteins for imaging, demonstrating that redesigned variants can function in real biological environments rather than only in silico.

Raygun was further evaluated with enzymes essential for cellular tasks. In these tests, engineered proteins maintained activity consistent with the models’ structural expectations, strengthening confidence that the framework preserves mechanistic compatibility.

Finally, receptor-binding experiments targeted epidermal growth factor (EGF), a clinically relevant signaling protein involved in wound healing and in pathways often dysregulated in cancer. Variants designed by Raygun produced validated receptor interactions, suggesting the approach can reach biologically meaningful endpoints.

Taken together, the study suggests Raygun could accelerate protein engineering for biotechnology and biomedical research, potentially supporting future development of more efficient therapeutic proteins and improved tools for gene therapy.

Subject of Research: Cells
Article Title: Miniaturizing and modifying natural proteins with Raygun
News Publication Date: 29-Jul-2026
Web References: https://www.nature.com/articles/s41586-026-10842-8
References: 10.1038/s41586-026-10842-8
Image Credits: Eamon Queeney / Duke University School of Medicine
Keywords: Protein design; Artificial intelligence

Tags: AI-driven functional protein designAI-powered protein redesigndeep learning in structural biologyinnovative protein engineering toolsmolecular engineering with AInatural evolution simulationprotein language modelsreliable protein variant generationsequence-to-structure translationsize-controlled protein engineeringstandardized protein representationtargeted protein modification