Scientists have succeeded in designing a new protein, which would take 500 million years to evolve in nature, in just a few hours using artificial intelligence. Artificial intelligence named this protein esmGFP. This development revolutionises biotechnology. In the research utilising the artificial intelligence model, an advanced language model called ESM3 was used to accelerate protein evolution. While normally in natural evolution, proteins are shaped by random mutations and natural selection, the ESM3 model was trained with data from 2.78 billion known proteins and generated entirely new amino acid sequences from scratch.

Artificial intelligence completed 500 million years of evolution in just hours

This newly created esmGFP protein is only 58% similar to the closest green fluorescent protein (GFP) that exists in nature. ESM3 identified 96 different mutations when designing this protein. If these mutations had accumulated randomly in natural evolution, it would have taken about half a billion years. But thanks to artificial intelligence, this process was completed in hours. Glowing proteins are found in nature in some marine organisms such as jellyfish and corals, and scientists use such proteins as microscopic biomarkers. The development of esmGFP has important potential in biological sciences as well as in medicine and environmental sciences.

Alex Rives, one of the leading figures in the research, states that the artificial intelligence model they have developed has the capacity to create functional proteins without being bound by the constraints of natural evolution. Rives and his team first developed the predecessors of ESM3 while working at Meta, and then furthered this research by founding the company EvolutionaryScale. While this development has the potential to revolutionise biology and medicine, it could also be used in environmental technologies such as carbon capture.