AI Breakthrough Simulates 500 Million Years of Protein Evolution
A groundbreaking study published in the journal Science highlights a significant advancement in the field of biology, where artificial intelligence (AI) has been utilized to simulate alternative pathways of protein evolution. The research, conducted by the startup EvolutionaryScale, established by former Meta researchers, utilizes a generative language model known as ESM3 to create new proteins, including a novel green fluorescent protein (GFP) that possesses only 58% similarity to existing versions.
ESM3 has been trained using over 771 billion data packets, converting vast biological data into a "language" that the AI model can interpret and manipulate. This unprecedented computational power opens new avenues in protein design and could lead to the development of innovative therapeutics and environmental applications.
Using this AI technology, the researchers were able to create a protein, named esmGFP, simulating what could have occurred in the evolutionary history of life on Earth. The findings support ongoing debates about evolutionary contingency, a theory popularized by evolutionary biologist Stephen Jay Gould, positing that small changes in early evolutionary events could lead to drastically different outcomes.
Experts believe this AI tool will provide deeper insights into the potential paths evolution could have taken, suggesting new biological possibilities that have not yet emerged naturally. This research marks a milestone in the intersection of AI and biology, promising future breakthroughs in understanding life’s complexities.
Weekly Newsletter
News summary by melangenews