mdtraj | open library for the analysis of molecular dynamics

I recommend you start reading the documentation, especially the "learn PyEMMA" section containing Jupyter notebooks teaching you the work-flow to extract properly weighted "pseudo" free-energy surfaces. Usually these surfaces are drawn into the dimensions of the first two slowest dynamical processes, but you can think of any other combination as well. These dimensions are defined by a TICA or VAMP projection, which are basically methods to extract the slow modes from your data, in case of proteins this contains folding and rare events.

As a primer I suggest reading this tutorial first, as it gives you a brief overview how to load and process your data to extract the slow modes. Note that this not yet contain Markov state modelling, so read further in the other examples to learn about that.