Raygun AI shrinks proteins by up to 50% while preserving function, Duke researchers report
This digest was compiled by AI from multiple sources — links to the originals are below.
Duke University researchers have developed Raygun, a generative AI framework that can miniaturize proteins by 10–25%—and in some cases more than 50%—without compromising their structure or function. The tool, described today in Nature, can also expand proteins beyond their natural size and introduce extensive sequence diversity, a capability that could streamline drug delivery and protein engineering. The method succeeds even as existing computational tools struggle with large-scale modifications involving both substitutions and insertions/deletions.
The Raygun Framework
Raygun encodes proteins not as variable-length sequences but as probability distributions in a fixed-dimensional space, making proteins of any length directly comparable. This encoding, derived from protein language model embeddings, allows the AI to coordinate substitutions, insertions, and deletions across the entire sequence. The framework is governed by just two parameters that control the extent of amino acid substitutions and length changes. By tuning these, researchers can shrink proteins by 10–25% on average, with some targets reduced by more than 50%, or expand proteins beyond their natural size.
Laboratory Validation
The Duke team tested Raygun on several proteins. They miniaturized fluorescent proteins to sizes shorter than 96% of the over 1,000 fluorescent proteins catalogued in FPbase, with two variants setting new compactness records. Raygun also produced a compact version of TurboID, a widely used biotin ligase for proteomics, preserving its enzymatic activity. In a demonstration of expansion, the tool generated epidermal growth factor (EGF) variants with higher binding affinity to the EGFR receptor than the wild-type protein, as confirmed by cell-based assays.
Overcoming Design Limitations
Current template-based protein design methods rely heavily on substitutions and become computationally infeasible when changes exceed a few dozen sites. Raygun’s ability to handle insertions and deletions—indels—mimics the coordinated mutations seen in natural evolution, acting, as Phys.org described it, like a ‘molecular shrink ray.’ This allows for large-scale modifications without losing the protein’s core structural integrity, a process the researchers liken to renovating a building while adding or removing entire rooms, not just repainting walls.
What's Next
The next step for the Duke team is to apply Raygun to a broader range of therapeutic proteins, including those used in gene therapy where compact size is critical. However, it remains unclear how these redesigned proteins will perform in living organisms, as in vivo stability and potential immune responses have yet to be tested.
4 sources
Raygun AI shrinks proteins by up to 50% while preserving function, Duke researchers report
