PDS-Lap is a machine learning methods which combines post-double-selection for baseline variable selection with Laplacian regularization. This approach is designed for the estimation and inference of drug effect comparions, incorporating external drug knowledge by integrating drug similarity information.