Test of Normality for Integrated Change Point Detection and Mixture Modeling

Single-molecule data often show step-like changes in the quantity measured between constant levels. Analysis of this data consists of detecting the steps, i.e., change point detection (CPD), and determining the levels, i.e., clustering. We describe a novel algorithm which integrates these two analyses, based on a statistical test of a normal distribution. The test of normality (TON) algorithm integrates statistical CPD with gaussian mixture model clustering. We used TON with both simulated data and ion channel patch-clamp recordings. It performed well with simulated data except at a high signal-to-noise ratio and when the frequency of steps was high compared to the sampling frequency. TON has advantages over separate CPD and mixture modeling algorithms, especially for complex single-molecule data. This was illustrated by its application to the maxichannel, an ion channel with multiple subconductance states.