Clustering of molecular dynamics trajectories via peak-picking in multidimensional PCA-derived distributions

TL;DR

This paper introduces a fast, automated clustering method for molecular dynamics trajectories using peak-picking in PCA-derived distributions, enabling efficient analysis of large simulation datasets.

Contribution

The authors present a novel, automated peak-picking clustering algorithm for MD trajectories that is robust, memory-efficient, and applicable to multidimensional PCA data.

Findings

Successfully clusters millions of structures from MD simulations.

Automates threshold determination using variance-explained criterion.

Implemented in analysis programs carma, grcarma, and cluster5D.

Abstract

We describe a robust, fast, and memory-efficient procedure that can cluster millions of structures derived from molecular dynamics simulations. The essence of the method is based on a peak-picking algorithm applied to three- and five-dimensional distributions of the principal components derived from the trajectory and automatically supports both Cartesian and dihedral PCA-based clustering. The density threshold required for identifying isolated peaks (which correspond to discrete clusters) is determined through the application of a variance-explained criterion which allows for a completely automated clustering procedure with no user intervention. In this communication we describe the algorithm and present some of the results obtained from the application of the method as implemented in the molecular dynamics analysis programs carma, grcarma. and cluster5D. We conclude with a discussion of the limitations and possible pitfalls of this method.