Commonly used methods for molecular representations in existing literature. All of them fail to achieve accurate molecular identity and feature extraction due to the intrinsic limitation of their learning principles, which motivates researchers to explore a new type of representation.