AI creates glowing protein that would've taken nature 500 million years to evolve

Fast-forwarding evolution

Serving tech enthusiasts for over 25 years.
TechSpot means tech analysis and advice you can trust.

What just happened? Scientists have used AI to design the blueprints for an entirely new protein that has never existed in nature. This AI-generated protein, dubbed esmGFP, would have taken half a billion years to evolve naturally. And the best part? It glows.

In a study published in Science, researchers detailed how they used advanced language models to fast-forward evolution, simulating hundreds of millions of years of genetic changes in just hours. The result? A synthetic version of green fluorescent protein (GFP) with an amino acid sequence only 58 percent similar to its closest natural counterpart.

For the uninitiated, GFPs are biomolecules that give certain marine creatures – like jellyfish – their vivid glow. Scientists frequently use them as biomarkers, attaching their genes to other proteins of interest to make them fluoresce under a microscope.

In nature, these glowing proteins evolved over eons through random genetic mutations. But the AI model behind this breakthrough, called ESM3, took a radically different approach. Instead of evolving proteins step by step like life on Earth, it was trained on a dataset of 2.78 billion known proteins – using one trillion teraflops of computing power – to generate entirely new hypothetical sequences.

For esmGFP specifically, the AI coded 96 mutations that would take over 500 million years to naturally arise in organisms like jellyfish or corals.

Alex Rives, co-founder of EvolutionaryScale, told Live Science that by inferring the fundamental biological rules, their model can create functional proteins that defy the constraints of natural evolution. Rives and his colleagues previously worked on precursor models to ESM3 at Meta before founding EvolutionaryScale last year. Just months later, the startup raised $142 million to advance its research.

// Related Stories

• This bird-inspired drone is more energy efficient and proficient at complex flight maneuvers
• Researchers successfully train robots to perform surgery by watching videos

However, not everyone is entirely convinced. Tiffany Taylor, an evolutionary biologist at the University of Bath, acknowledged to Live Science that the model holds promise for drug development and bioengineering. Still, she cautioned that AI protein models don't account for the complex selective forces shaping entire organisms.

Despite these concerns, the study highlights how AI could dramatically expand the range of synthetic proteins available, with potential applications in medicine and environmental science.

"The model has the potential to accelerate discovery across a broad range of applications, ranging from the development of new cancer treatments to creating proteins that could help capture carbon," a press release from last year noted.