Abstract

To give meaning to biomolecular processes observed in simulation, we need to characterize the conformational ensemble in terms of transitions between metastable states. Extensive work has been done to discover the metastable states through clustering. Most recent approaches perform dimensionality reduction on a large set of internal coordinates, followed by a clustering technique. In recent work, we demonstrated that high quality clustering can be performed through Gaussian Mixture Models directly on positions through a maximum likelihood approach, if structural alignment is performed in an iterative fashion that takes into account covariances between residue positions. Following these developments, we demonstrate that dimensionality reduction performed on positions after this procedure is a promising technique for developing new reaction coordinates between states.