Protein map unifies sequence and structure for evolution insights
Summary and headline written by AI from the source article. How we work
An international team of scientists from institutions including the Earth-Life Science Institute (ELSI) in Tokyo and universities in Israel developed a new protein language model that combines information from protein amino acid sequences and their three-dimensional structures.
Researchers have long studied proteins, the molecules responsible for nearly all functions within cells, by grouping them based on sequence similarities, similar to how species are classified. However, recent advances in artificial intelligence offered new methods for understanding relationships between proteins.
The team created a model called Contrastive Learning Sequence-Structure (CLSS) that generates a shared map showing how proteins relate to one another based on both their sequence and structure. CLSS uses a technique called contrastive learning to ensure that representations of a protein’s sequence and structure are located close together on the map.
When tested against other protein language models, CLSS successfully integrated both types of information and accurately reflected relationships found in existing protein classifications without being given that information directly. The model can even work with protein fragments, small pieces that have been reused throughout evolution, offering clues to ancient relationships. By displaying biological properties on the map, the researchers found patterns, such as proteins linked to organic cofactors clustering in specific areas.
The scientists expect this unified approach to aid in database searches, protein engineering, and understanding protein evolution over billions of years.
