AI Generates Fluorescent Protein That Nature Would Need 500 Million Years to Evolve

A novel fluorescent protein has been created using artificial intelligence (AI), with scientists estimating that its natural evolution would have required half a billion years. The protein, known as esmGFP, was designed by an AI model trained on extensive biological data, leading to the development of a structure distinct from naturally occurring green fluorescent proteins found in jellyfish and corals. The breakthrough is expected to contribute to advancements in medicine and protein engineering.

Study Reveals AI-Driven Molecular Evolution

According to the study published in Science, the AI model ESM3 was used to generate esmGFP by filling in missing genetic sequences based on data from 2.78 billion naturally occurring proteins. The result was a protein that shares only 58 percent of its sequence with the closest known equivalent, a human-modified protein derived from bubble-tip sea anemones (Entacmaea quadricolor). Scientists noted that 96 distinct genetic mutations would have been required for esmGFP to evolve naturally, a process estimated to take over 500 million years.

How the AI Model Works

The AI model, developed by researchers at EvolutionaryScale, functions by predicting and completing protein sequences using language-modeling techniques similar to those used in text-based AI systems. Unlike traditional evolution, where proteins undergo gradual changes through natural selection, ESM3 generates functional proteins by exploring vast possible genetic variations. Speaking to Live Science, Alex Rives, co-founder and chief scientist at EvolutionaryScale, stated that the AI system learns fundamental biological principles and can create functional proteins beyond the constraints of natural evolution.

Applications in Biotechnology

Green fluorescent proteins are widely used in research laboratories, often attached to other proteins to track cellular processes. The study’s findings suggest that AI-driven protein engineering could accelerate drug development and other applications in biotechnology. Tiffany Taylor, an evolutionary biologist at the University of Bath, noted in her analysis of the preprint study that while AI models like ESM3 offer new possibilities in protein design, the broader complexities of natural selection should not be overlooked.