Collective coordinates, as obtained by a principal component analysis of atomic fluctuations, are commonly used to predict a low-dimensional subspace in which essential protein motion is expected to take place. The definition of such an essential subspace allows to characterize protein functional, and folding, motion, to provide insight into the (free) energy landscape, and to enhance conformational sampling in molecular dynamics simulations. Here, we provide an overview on the topic, giving particular attention to some methodological aspects, such as the problem of convergence, and mentioning possible new developments.