AI model integrates protein sequences and structures to map evolution

An international research team has developed a new protein language model designed to integrate amino acid sequences with three-dimensional structures. This AI-driven approach allows scientists to map relationships across the protein universe and study evolutionary patterns spanning billions of years.

By converting proteins into numerical representations known as embeddings, the model enables researchers to visualize a ‘protein world map’ where proteins with similar properties are grouped together. This method aims to address fundamental questions in evolutionary biochemistry regarding how protein families are related and how they originated.

The research involved contributors from the Earth-Life Science Institute (ELSI) at the Institute of Science Tokyo, the University of Haifa, and Tel Aviv University. The findings were published in the Proceedings of the National Academy of Sciences (PNAS).