FIG. 2.

GMRQ scores for two-time scale MSMs generated for $\lambda$-repressor (PDB ID: 1lmb) containing varying numbers of microstates show that the cross-validation is necessary to describe a system’s underlying dynamics. All models were constructed using k-centers clustering with the RMSD distance metric. The error bars signify the score standard deviation generated from five cross-validation iterations (the error bars on the training scores are negligibly small). The discrepancy between training and test scores (light blue and dark blue, respectively) as microstate number increases is likely due to overfitting during training; thus these models exhibit poorer performance on data that were hidden from the fitting process.