Researchers have developed ESM3, an advanced generative language model for proteins that can simulate evolutionary processes to create functional proteins vastly different from known ones. ESM3 integrates sequence, structure, and function to generate proteins following complex prompts, such as a new fluorescent protein called esmGFP. This breakthrough demonstrates ESM3’s potential in protein engineering, offering creative solutions to biological challenges.
ESM3 is a sophisticated generative language model that can predict and generate protein sequences, structures, and functions effectively. It processes these aspects through transformer blocks with geometric attention, trained on vast datasets to scale up to 98 billion parameters. The model’s generative capabilities allow it to create diverse, high-quality proteins that differ significantly from known natural proteins, offering advanced, programmable protein design options beyond natural evolutionary patterns.
Scaling and fine-tuning ESM3 models significantly enhance their ability to generate proteins aligned with complex prompts, showing improved performance when aligned with specific objectives. By prompting the model with critical residues and structures necessary for GFP function, ESM3 created a unique fluorescent protein, esmGFP, significantly different from known proteins. This process effectively simulates millions of years of evolutionary potential in generating new functional proteins.
ESM3’s versatility in following prompts and generating novel protein designs showcases its potential in protein engineering and exploration beyond natural evolutionary patterns. The model’s ability to adapt to challenging tasks and generate accurate protein structures demonstrates its effectiveness in creating diverse and high-quality proteins, offering innovative solutions to biological challenges.
Source link
Source link: https://www.marktechpost.com/2024/06/26/evolutionaryscale-introduces-esm3-a-frontier-multimodal-generative-language-model-that-reasons-over-the-sequence-structure-and-function-of-proteins/?amp