We describe a protocol for prediction of ligand crystallographic binding poses and for effective analysis of protein-ligand dynamics in mechanistic or drug design studies of biomolecules. Our protocol includes guidelines for initial setup of simulations, approaches for performing the simulations that allow for diverse computing architectures, a metric for evaluating convergence and “when to stop” trajectories, and a model for efficiently aggregating and analyzing the simulation data. We aggregate dynamic ligand binding information with a statistical Markov State Model to correctly predict four FK506-derived ligand binding poses to the immunophilin protein FKBP12, with our predictions in agreement with available experimental structures with < 3 A RMSD. We also give a blind prediction for a distinct,steroidal ligand, that is also validated with experimental information. We assert that our approach can give accurate binding pose predictions at a timeline of 1-3 months on state-of-the-art resources, including GPU clusters, supercomputers, or cloud computing, for quicker results in cases where experimental structures are difficult or intractable to produce.