Title: Generalized Coordinates from Molecular Dynamics Simulations: Reaction Pathways and Dimensionality Reduction
Speaker: Benjamin BOUVIER
Theoretical and Computational Biophysics Department
Max Planck Institute for Biophysical Chemistry
Goettingen, Germany
Abstract Progresses in computer hardware and software allow all-atom molecular dynamics (MD) simulations to be applied to increasingly large and complex biological systems. However, most problems of interest (conformational transitions in macromolecules, protein folding, etc^E) occur on timescales which cannot be accurately sampled by such methods, due to the very large number of degrees of freedom involved. Reducing the representation of the system to a few relevant generalized coordinates which capture the properties of interest thus remains highly desirable. Exploring the conformational subspace spanned by the reduced coordinates provides insight into conformations that would be sampled at much longer simulation times. Sampling can also be selectively enhanced along one or several of the coordinates, accelerating the associated transition to simulation-accessible timescales.

However, finding a relevant set of global coordinates is usually not a trivial task. Two examples are presented in this talk. The flipping of a base in DNA, a ubiquitous process occurring on the millisecond timescale, is found to be best described using a generalized reaction coordinate extracted from a principal component analysis. Sampling along this coordinate is enhanced using the conformational flooding technique, yielding the pathway and associated free energy profile. By combining competitive learning and topology-preserving mapping techniques, we attempt to provide optimal reduced-dimensional representations of all-atom MD simulations of small protein systems. These retain the salient features contained in the MD trajectories, facilitating classification and prediction of the possible conformations of the macromolecules.