Researchers at EvolutionaryScale have made a groundbreaking discovery by using artificial intelligence to generate a new fluorescent protein, esmGFP, which could revolutionize biochemistry. This achievement, announced on January 30, 2025, marks a significant leap in protein engineering, simulating what would have taken nature 500 million years to evolve.
Proteins, essential for various biological functions, are composed of sequences of amino acids. The ESM3 model, a multimodal generative AI, was trained on an extensive dataset of 771 billion tokens derived from natural and synthetic protein sequences. This advanced model not only understands the fundamental language of protein biology but also generates entirely new sequences.
In a demonstration of its capabilities, ESM3 produced a variant of the green fluorescent protein (GFP), commonly used in cell and molecular biology. The new protein shares only 53% sequence similarity with its closest natural counterpart, suggesting a vast evolutionary divergence. According to Alexander Rives, co-founder and chief scientist at EvolutionaryScale, this discovery could lead to innovative applications in medicine and environmental science.
Potential uses for the generated proteins include designing enzymes to break down plastic waste and developing new therapeutic agents. While the implications of this research are promising, experts caution against overestimating the capabilities of AI in replicating the complexities of natural evolution. Nonetheless, the findings present exciting possibilities for future scientific exploration.