Robust Hypothesis Tests for Detecting Statistical Evidence of 2D and 3D Interactions in Single-Molecule Measurements

Abstract

A variety of experimental techniques have improved the 2D and 3D spatial resolution that can be extracted from \emph{in vivo} single-molecule measurements. This enables researchers to quantitatively infer the magnitude and directionality of forces experienced by biomolecules in their native cellular environments. Situations where such forces are biologically relevant range from mitosis to directed transport of protein cargo along cytoskeletal structures. Models commonly applied to quantify single-molecule dynamics assume that effective forces and velocity in the $x,y$ (or $x,y,z$) directions are statistically independent, but this assumption is physically unrealistic in many situations. We present a hypothesis testing approach capable of determining if there is evidence of statistical dependence between positional coordinates in experimentally measured trajectories; if the hypothesis of independence between spatial coordinates is rejected, then a new model accounting for 2D (3D) interactions should be considered to more faithfully represent the underlying experimental kinetics. The technique is robust in the sense that 2D (3D) interactions can be detected via statistical hypothesis testing even if there is substantial inconsistency between the physical particle's actual noise sources and the simplified model's assumed noise structure. For example, 2D (3D) interactions can be reliably detected even if the researcher assumes normal diffusion, but the experimental data experiences "anomalous diffusion" and/or is subjected to a measurement noise characterized by a distribution differing from that assumed by the fitted model. The approach is demonstrated on control simulations and on experimental data (IFT88 directed transport in the primary cilium).