Supporting Materials

SI methods 1. Convergence analysis Both the RMS average correlations (RAC) and Kullback–Leibler divergence (KLD) methods [18-19] were used to access whether the individual simulations are reasonably converged. The RAC functionality can be thought of as a pseudo-autocorrelation function for RMSD values. This essentially measures the convergence of the overall average structure at different time intervals within a single trajectory. For a given time interval or lag (τ) a straight coordinate running average over that time interval is performed over the entire trajectory [24]; each sliding averaged structure over the time interval τ is then either fit to the first averaged structure (time 0 − τ) or a reference structure specified by the user, and finally the average RMSD value of all averaged structures of length τ is calculated according to: