MDAnalysis: Dimension Reduction

Dimension reduction of a molecular dynamics (MD) simulation is meant to provide insight into the slow underlying dynamical motions of a set of molecules. Dimension reduction algorithms are either nonlinear or linear transformations, both categories providing distinct benefits and shortcomings. Existing implementations of any dimension reduction algorithm work by decomposing either some subset of an N-by-P “observation matrix” delineating the N “observations” of a system with P coordinates or with a P-by-P correlation matrix. Functions in MDAnalysis avoid creating such a matrix as it would proscribe users from analyzing very large MD simulations on a workstation or laptop. The goal of this project will be to implement Principal Component Analysis and diffusion maps in MDAnalysis such that both methods are accessible to aforementioned users.