Combination of density-clustering and supervised classification for event identification in single-molecule force spectroscopy data

Abstract

Single-molecule force spectroscopy (SMFS) measurements of the dynamics of biomolecules typically require identifying massive events and states from large data sets, such as extracting rupture forces from force-extension curves (FECs) in pulling experiments and identifying states from extension-time trajectories (ETTs) in force-clamp experiments. The former is often accomplished manually and hence is time-consuming and laborious while the latter is always impeded by the presence of baseline drift. In this study, we attempt to accurately and automatically identify the events and states from SMFS experiments with a machine learning approach, which combines clustering and classification for event identification of SMFS (ACCESS). As demonstrated by analysis of a series of data sets, ACCESS can extract the rupture forces from FECs containing multiple unfolding steps and classify the rupture forces into the corresponding conformational transitions. Moreover, ACCESS successfully identifies the unfolded and folded states even though the ETTs display severe nonmonotonic baseline drift. Besides, ACCESS is straightforward in use as it requires only three easy-to-interpret parameters. As such, we anticipate that ACCESS will be a useful, easy-to-implement and high-performance tool for event and state identification across a range of single-molecule experiments.