Raygun: Duke Researchers Develop AI Tool That Can Shrink and Supersize Proteins While Preserving Function
核心洞察
Duke University researchers developed Raygun (搜索), an AI framework that can resize proteins—making them shorter or longer—while preserving their structure and function, published in Nature.
Raygun (搜索) builds on protein language models like ESM-2 (搜索), converting proteins into standardized mathematical representations to enable extensive modifications without disrupting structural integrity.
The tool requires only two user settings: one controlling how much the protein sequence changes and another determining whether the protein becomes shorter or longer.
Researchers at Duke University School of Medicine have unveiled an artificial intelligence framework capable of redesigning proteins at a scale previously observed only in natural evolution. The tool, called Raygun (搜索), can shrink, expand, or otherwise extensively modify existing proteins while preserving their structure and biological function—a capability that has long eluded scientists working in protein engineering.
The findings were published today in Nature.
"There is a language that governs how a protein's amino acid sequence gives rise to its shape and function, but scientists don't fully understand that language," said Rohit Singh, PhD, co-corresponding author of the study and an assistant professor of biostatistics and bioinformatics and cell biology at Duke. "Protein language models act as a kind of translator, learning patterns from millions of protein sequences and linking those patterns to biological structure and function."
A New Mathematical Approach to Protein Representation
Rather than designing proteins from scratch—an approach taken by many existing AI protein-building models—Raygun (搜索) builds on protein language models to make extensive modifications to existing proteins. The method leverages ESM-2 (搜索), a large language model trained on millions of natural protein sequences, but introduces a fundamentally new way of representing proteins.
Most protein language models create representations of proteins as amino-acid sequences of varying lengths, a system that makes it difficult to generate proteins that maintain their overall structure and function at different sizes. Raygun (搜索) takes a different approach: it divides each protein into pieces and translates the information in each piece into numerical patterns. The tool then uses these patterns to create a standardized version of the protein that follows particular rules. Because proteins of any size can be represented in this same format, Raygun can compare, shrink, expand, or otherwise redesign proteins while preserving important structural and functional features.
The model learns those rules, allowing it to generate the protein at various sizes without compromising its structural integrity.
Simplicity in Design
Researchers can steer Raygun (搜索) using only two settings: one that determines how much a protein's sequence changes and another that controls whether the protein becomes shorter or longer. This streamlined interface belies the sophisticated computational work happening beneath the surface, where the system performs insertions, deletions, and substitutions of single protein subunits—mirroring some of the same steps used in natural evolution.
"Sometimes a smaller protein can do things or fit into places where a larger protein cannot," Singh noted, highlighting the practical motivation behind the technology.
Expert Reception and Future Applications
Fajie Yuan, a computational biologist who works on protein language models at Westlake University in Hangzhou, China, who was not involved in the work, described Raygun (搜索) as a "meaningful and creative" first step towards editing existing proteins.
"For many real applications, that is exactly what researchers want—not a completely new protein, but a better, smaller, larger or more adaptable version of one they already trust," Yuan said.
The development opens new possibilities for engineering proteins with customized sizes and properties for use in biological research and, ultimately, as targeted medicines. Protein-building models have already advanced to the point where they can design antibodies with clinical potential, and the ability to resize natural proteins represents a tantalizing prospect for biotechnology applications. Efforts to use AI systems to modify existing proteins have historically been limited because such modifications tend to disrupt structure or function—a challenge that Raygun (搜索) appears to address.