A structure-aware generative AI framework for revealing functional relationships in protein families.

A structure-aware generative AI framework reveals functional relationships in protein families by integrating sequence and structural data, identifying conserved and variable regions.

• Why it matters: Understanding how sequence variations relate to structural and functional differences is crucial for deciphering protein evolution and designing new proteins, but current methods lack a unified quantitative approach.
• What they did: The framework constructs parallel sequence and structure-informed representations using variational autoencoders and coevolution analysis across five protein families, quantifying their relationships with information-theoretic metrics.
• The result: It uncovers conserved structural scaffolds, functional subfamilies, and hidden evolutionary links, enabling mechanistic insights and guiding protein design by sampling near functional regions.