AI Simulates 500 Million Years of Evolution to Innovate Protein Design
A groundbreaking study published in the journal Science highlights how artificial intelligence (AI) is reshaping the field of biology. Researchers from the startup EvolutionaryScale, comprised of former Meta employees, utilized advanced computational resources to simulate an alternative evolutionary path for proteins. The AI system, called ESM3, generated a new green fluorescent protein, known as esmGFP, which demonstrated only 58% similarity to existing versions. This innovation suggests an evolutionary possibility that nature did not explore over the course of 500 million years.
The ESM3 model is a product of extensive training, leveraging over 771 billion data packets sourced from a vast array of existing protein sequences, structures, and functional traits. By harnessing this data, ESM3 can "speak" the biological language of proteins and aid in the design of new proteins for therapeutic and environmental purposes.
This research addresses a long-standing debate in evolutionary biology regarding whether the course of evolution is predetermined or contingent. By showcasing viable protein structures that have never existed, the study supports the idea that many biological paths could have emerged differently. Experts in the field affirm that while deterministic factors exist in evolution, AI technology can provide insight into the vast realm of biological possibilities that could have developed under different circumstances.
Weekly Newsletter
News summary by melangenews