AI Unveils 500 Million Years of Evolutionary Possibilities in Protein Design
A new study published in the journal Science highlights groundbreaking advances made by EvolutionaryScale, an AI startup founded by former Meta researchers. Utilizing the EvolutionaryScale Model 3 (ESM3), the team simulated 500 million years of evolution to create a novel green fluorescent protein, termed esmGFP. This protein showcases only 58% similarity to the most closely related naturally occurring counterpart, revealing significant alternative evolutionary pathways.
The ESM3 model is a sophisticated generative language model capable of interpreting and synthesizing biological data from extensive protein sequence, structure, and functional information. Such revolutionary technology has the potential to inform future developments in therapeutic applications and environmental solutions, ultimately reshaping our understanding of protein biology.
Alexander Rives, chief scientist of EvolutionaryScale, emphasized that ESM3 can generate proteins from first principles, akin to constructing machines or microchips. This innovation opens a new realm of biologically viable possibilities, suggesting paths that evolution could have taken but did not.
Experts in the field, including Jonathan Losos and Zachary Blount, underscore the implications of this research for the concept of evolution, corroborating the idea that alternative evolutionary trajectories could have led to diverse biological outcomes. As this technology becomes available for broader scientific exploration, it promises to enhance the understanding of evolutionary dynamics and protein functionality.
Weekly Newsletter
News summary by melangenews