A recent study published in Science highlights a significant breakthrough in molecular biology, where an AI model successfully simulated half a billion years of molecular evolution to create a new protein, named esmGFP.
This glowing protein resembles those found in jellyfish and corals and could contribute to the development of new drugs. Currently, it exists only as a digital code, but scientists assert it contains a genetic blueprint for a novel type of green fluorescent protein.
The research reveals that the amino acid sequence of esmGFP shares only 58% similarity with the closest known protein, a modified human version of a protein found in sea anemones. The study suggests that nature would have required 96 different genetic mutations over 500 million years to develop such a protein.
The AI model, ESM3, operates without the constraints of natural evolution, filling gaps in protein sequences based on knowledge of billions of possible evolutionary pathways. Trained on data from 2.78 billion natural proteins, it generates new protein sequences not previously found in living organisms.
This advancement marks a leap in protein engineering, particularly as fluorescent proteins are widely used in biological research to monitor cellular processes and molecular interactions. However, some scientists caution that while powerful, this method cannot replace the complex natural processes that have shaped life over millions of years.
The research was developed by EvolutionaryScale, stemming from work initiated by researchers at Meta, the parent company of Facebook and Instagram, before founding EvolutionaryScale in 2024. The study has been peer-reviewed, enhancing its credibility and paving the way for further research on AI's role in protein development and future medical therapies.