FoldToken2: Learning compact, invariant and generative pr... | ResearchHub