Function replace_rare_residues 

pub fn replace_rare_residues(sequence: &str) -> String

Rostlab’s [UZOB] -> X convention, for callers reproducing published ProtT5 embeddings.

Offered rather than applied: see the module docs. Selenocysteine (U) and pyrrolysine (O) are real residues with their own learned embeddings, so collapsing them loses information that the model actually has.