Universal Descriptors of 'quasi Transition States' for Small-Data-Driven Asymmetric Catalysis Prediction in Machine Learning Model