A recent study has revealed that an artificial intelligence (AI) model simulated 500 million years of molecular evolution to design the code for a new and previously unknown protein. This glowing protein, named esmGFP, resembles those found in jellyfish and corals and could aid in developing new medications. Proteins, essential for various bodily functions including muscle building and disease fighting, are made up of amino acids whose sequences are dictated by genes. The esmGFP protein exists only in computer code and represents a unique type of green fluorescent protein, with only 58% sequence similarity to the closest known fluorescent protein—a modified version of one from bubble-tip sea anemones. Creating esmGFP naturally would require 96 genetic mutations over millions of years. The AI model, called ESM3, was introduced by EvolutionaryScale and has been validated by independent scientists.

Unlike traditional evolutionary processes, ESM3 fills in incomplete protein sequences, leveraging insights from a vast dataset of 2. 78 billion natural proteins. This gap-filling approach helps generate potential protein structures that evolution might not have produced. Researchers trained ESM3 using properties of proteins, allowing it to predict and complete protein codes like a person completing sentences. The implications of this research extend to various applications in protein engineering, including drug design. However, while experts like Tiffany Taylor from the University of Bath acknowledge the potential of AI in protein engineering, she cautions that such models might oversimplify the complex processes shaped by millions of years of natural selection.