Related Experiment Video
Updated: Aug 5, 2026
08:34
OaAEP1-Mediated Enzymatic Synthesis and Immobilization of Polymerized Protein for Single-Molecule Force Spectroscopy
Published on: February 5, 2020
Miniaturizing and modifying natural proteins with Raygun
Kapil Devkota1, Daichi Shonai2, Joey Mao2
1Department of Biostatistics and Bioinformatics, Duke University, Durham, NC, USA.
Nature
|July 29, 2026
Summary
Raygun, a new AI framework, engineers proteins by encoding them as probability distributions. This enables function-preserving miniaturization, expansion, and modification of proteins, mimicking natural evolution.
Area of Science:
- Computational biology
- Protein engineering
- Artificial intelligence in life sciences
Background:
- Protein engineering struggles to replicate natural evolution's sequence modification capabilities.
- Existing protein language models provide representations but lack tools for large-scale, function-preserving modifications.
Purpose of the Study:
- Introduce Raygun, a generative AI framework for protein engineering.
- Enable protein miniaturization, modification, and augmentation.
- Develop a method for length-agnostic protein representation.
Main Methods:
- Developed Raygun, a generative AI framework using probabilistic encoding of protein sequences from language model embeddings.
- Encoded proteins as probability distributions in fixed dimensions for length commensurability.
- Controlled protein modifications (substitutions, length changes) using two parameters.
Main Results:
- Raygun achieved protein miniaturization (10-25%, up to 50%), expansion, and sequence diversity while preserving structure and function.
- Validated Raygun by miniaturizing fluorescent proteins and TurboID, and expanding epidermal growth factor (EGF).
- Generated EGF variants with enhanced EGFR-binding affinity.
Conclusions:
- Protein function can be captured in a length-agnostic representation.
- Raygun facilitates large-scale, coordinated protein sequence modifications akin to natural evolution.
- This framework advances computational protein design and engineering capabilities.
