Dimension reduction

A molecular dynamics trajectory with \(N\) atoms can be considered a path through \(3N\)-dimensional molecular configuration space. It remains difficult to extract important dynamics or compare trajectory similarity from such a high-dimensional space. However, collective motions and physically relevant states can often be effectively described with low-dimensional representations of the conformational space explored over the trajectory. MDAnalysis implements two methods for dimensionality reduction.

Principal component analysis is a common linear dimensionality reduction technique that maps the coordinates in each frame of your trajectory to a linear combination of orthogonal vectors. The vectors are called principal components, and they are ordered such that the first principal component accounts for the most variance in the original data (i.e. the largest uncorrelated motion in your trajectory), and each successive component accounts for less and less variance. Trajectory coordinates can be transformed onto a lower-dimensional space (essential subspace) constructed from these principal components in order to compare conformations. Your trajectory can also be projected onto each principal component in order to visualise the motion described by that component.

Diffusion maps are a non-linear dimensionality reduction technique that embeds the coordinates of each frame onto a lower-dimensional space, such that the distance between each frame in the lower-dimensional space represents their “diffusion distance”, or similarity. It integrates local information about the similarity of each point to its neighours, into a global geometry of the intrinsic manifold. This means that this technique is not suitable for trajectories where the transitions between conformational states is not well-sampled (e.g. replica exchange simulations), as the regions may become disconnected and a meaningful global geometry cannot be approximated. Unlike PCA, there is no explicit mapping between the components of the lower-dimensional space and the original atomic coordinates; no physical interpretation of the eigenvectors is immediately available.

For computing similarity, see the tutorials in Trajectory similarity.