Text this: Contrastive-learning of language embedding and biological features for cross modality encoding and effector prediction