Sequential Change Point Detection in Molecular Dynamics Trajectories

Abstract

Motivated from a molecular dynamics context we propose a sequential change point detection algorithm for vector-valued autoregressive models based on Bayesian model selection. The algorithm does not rely on any sampling procedure or assumptions underlying the dynamics of the transitions and is designed to cope with high-dimensional data. We show the applicability of the algorithm on a time series obtained from numerical simulation of a penta-peptide molecule.