Molecular machine learning models often fail to generalize beyond the chemical space of their training data, limiting their ability to reliably perform predictions on structurally novel bioactive ...