Built a kinetic model for a ligand-protein binding process (~10TB time series data), combining clustering, principle component analysis and Markov modeling. Increased the accuracy of the binding rate calculation by 40%Applied long short-term memory recurrent neutral network to predict and generate molecular trajectories and saved 70% computational cost compared to only using molecular simulationsDiscovered an RPN11 inhibitor drug candidate by structure-based virtual screening in collaboration with a pharmacology lab