FIG. 3.

GMRQ scores for two-time scale MSMs generated for twelve ultra-long protein folding datasets¹⁸ where additional dimensionality reduction has been omitted (red), performed with PCA (green) or performed with tICA (blue) demonstrate that tICA systematically improves models. All MSMs were made using α-angle featurization and mini-batch k-means clustering. In all cases, the best tICA model performs better than or equivalently to both the best PCA model and the best model created directly from α-angle features. The error bars signify the score standard deviation generated from five cross-validation iterations. The number of amino acids of each protein is given below its PDB ID.