Learning the language of proteins

Abstract: By conceptualizing protein sequences as a language, where each amino acid represents a unique 'word,' we study the intricate 'grammar' that governs how protein sequences encode functional information. Our approaches leverage protein language models to study the syntax of protein sequences, enhancing our ability to predict and rationalize mutational effects. We specifically focus on the question of whether language models are useful tools to detect functional divergence in proteins, and show new ways that protein language models can be used as a basis for functional annotation of uncharacterized proteins.