Can principal components yield a dimension reduced description of protein dynamics on long time scales?

The suitability of principal component analysis (PCA) to yield slow collective coordinates for use within a dimension reduced description of conformational motions in proteins is evaluated. Two proteins are considered, T4 lysozyme and crambin. We present a quantitative evaluation of the convergence of conformational coordinates obtained with principal component analysis. Detailed analyses of (>200 ns) molecular dynamics trajectories and crystallographic data suggests that simulations of a few nanoseconds should generally provide a stable and statistically reliable definition of the essential and near constraints subspaces. Moreover, a systematic assessment of the density of states of the dynamics of all principal components showed that for an optimal separation of time scales it is crucial to include also side chain atoms in the PCA.