Excited to launch AutoStepfinder, a method that allows step detection in a wide variety of single-molecule trajectories. This project was a great collaboration with @jacobkers from the @cees_dekker lab during my time in the @ChirlminJoo_Lab.

https://www.cell.com/patterns/fulltext/S2666-3899(21)00082-9">https://www.cell.com/patterns/...

AutoStepfinder is the successor or Stepfinder and provides an automated step detection method that requires minimal knowledge of the location of steps and the various signal contributions in the data. In addition, we created a graphical user interface to facilitate the process.

AutoStepfinder iteratively performs two complementary fits with an equal number of steps. The fit is placed at locations that yield the biggest reduction in the variance and is compared to the variance of a worst-case-fit with the same number of steps, called a counter fit.

To assess the quality of the fit, algorithm generates a step spectrum that displays a sharp peak when the signal harbors step-like features, whereas smooth changes result in a flat curve. The quality is assessed over multiple rounds, allowing for step fitting at various scales.

We performed extensive benchmarking and show that AutoStepfinder performs robustly on a wide variety of signals and noise types.

Lastly, we have created several auxiliary tools to simulate data and allow the user to validate their findings.