Researchers have made a groundbreaking discovery in protein engineering using an AI model called ESM3. The team simulated 500 million years of evolution and created a novel fluorescent protein, revolutionizing the field.
ESM3 uses a multimodal generative language model to reason over protein sequence, structure, and function. This approach allows for the “searching” of potential proteins, enhancing our understanding of naturally evolved proteins and enabling the creation of new ones.
The training data for ESM3 consists of 771 billion unique tokens created from 3.15 billion protein sequences, 236 million protein structures, and 539 million proteins with function annotations. The model can train up to 98 billion parameters and is now available in public beta via an API.
This breakthrough enables scientists to engineer proteins programmatically or through interactive browser-based apps. Researchers can access the ESM3 model for free through its EvolutionaryScale Forge API or use the open model’s code and weights.
Source: https://scitechdaily.com/fast-forwarding-evolution-ai-mimics-500-million-years-of-biology