A team of scientists from the startup EvolutionaryScale has successfully created an artificial fluorescent protein using artificial intelligence (AI). The company’s ESM3 model, a generative language model, was trained on 771 billion data packets derived from existing proteins to create a new green fluorescent protein, esmGFP. This achievement is equivalent to simulating 500 million years of evolution and opens up new avenues for designing proteins with therapeutic functions, environmental remediation capabilities, and other potential applications.
The study, published in the journal Science, uses AI to recreate an evolutionary path that nature did not explore. The ESM3 model understands the biological language of proteins and creates new ones by feeding on sequence, structure, and function data from existing proteins. This is a significant step towards using AI as a tool for building from first principles in biology.
The creation of esmGFP surprised the researchers, who found that it was only 58% similar to the most closely related GFP. This highlights the possibility that evolution could have unfolded differently, and humans may not be the product of a predetermined path. The study supports the idea of contingency, which suggests that evolution is influenced by random events rather than a fixed course.
The discovery has sparked debate among scientists, with some arguing for determinism and others supporting contingency. However, researchers believe that AI can help us understand what is contingent and what is not, allowing us to explore biological possibilities and compare them to existing organisms.
As the field of AI in biology continues to evolve, this study marks an exciting step forward in our understanding of protein evolution and its potential applications. The availability of ESM3 as a new tool for designing proteins with therapeutic functions and other capabilities opens up new avenues for research and development.
Source: https://english.elpais.com/science-tech/2025-01-20/ai-simulates-500-million-years-of-evolution-to-discover-artificial-fluorescent-protein.html