AI Simulates 500 Million Years of Evolution, Discovering New Proteins
A new study published in the journal Science reveals that artificial intelligence (AI) has accomplished an unprecedented feat in biological research by simulating 500 million years of evolution to design a novel green fluorescent protein (GFP). This groundbreaking work was conducted by EvolutionaryScale, a startup founded by former Meta researchers.
The AI model, known as ESM3, functions similarly to language models like ChatGPT but focuses on generating proteins instead of text. The model utilizes extensive data—including 3.15 billion sequences and 236 million structures—to learn the language of proteins and develop entirely new variants. The resulting fluorescent protein, named esmGFP, displays only 58% similarity to existing GFPs previously found in nature.
This research provides insights into alternative evolutionary pathways that nature did not pursue. The study emphasizes the flexibility of evolutionary processes, aligning with the thoughts of evolutionary biologist Stephen Jay Gould, who argued that if evolution were replayed, its outcomes would likely differ. Experts in the field, such as Jonathan Losos and Zachary Blount, have noted that this AI-driven approach excels in exploring biological possibilities that have yet to manifest in the natural world.
As ESM3 becomes accessible to the scientific community, it represents a significant step in the potential applications for protein design, which could lead to breakthroughs in therapeutic and environmental solutions.
Weekly Newsletter
News summary by melangenews