Using Dimensionality Reduction to Better Capture RNA and Protein Folding Motions

Molecular motions, including both protein and RNA, play an essential role in many biochemical processes. Simulations have attempted to study these detailed large-scale molecular motions, but they are often limited by the expense of representing complex molecular structures. For example, enumerating all possible RNA conformations with valid contacts is an exponential endeavor, and the complexity of protein motion increases with the model’s detail and protein length. In this paper, we explore the use of dimensionality reduction techniques to better approximate protein and RNA motions. We present two new methods to study motions: (1) an evaluation technique to compare different distributions of conformations and (2) a way to identify likely local motion transitions. We combine these two methods in an existing motion framework to study large-scale motions for both proteins and RNA. We show that dimensionality reduction can be effectively applied, even to discrete conformation spaces (as for RNA secondary structure) that do not typically lend themselves to reduction techniques.