A Comparison of Step-Detection Methods: How Well Can You Do?
Brian C Carter 1, Michael Vershinin 2 and Steven P. Gross 3*
1 UC Irvine
2 UC Irvine
3 University of California, Irvine
* To whom correspondence should be addressed. E-mail: sgross{at}uci.edu.
Submitted on April 11, 2007
Revised on May 31, 2007
Accepted on 27 August 2007
 |
Abstract |
|---|
Many biological machines function in discrete steps, and detection of such steps can provide insight into the machines' dynamics. It is therefore crucial to develop an automated method to detect steps, and determine how its success is impaired by the significant noise usually present. A number of step detection methods have been used in previous studies, but their robustness and relative success rate have not been evaluated. Here, we compare the performance of four step detection methods on artificial benchmark data (simulating different data acquisition and stepping rates, as well as varying amounts of Gaussian noise). For each of the methods we investigate how to optimize performance both via parameter selection and via pre-filtering of the data. While our analysis reveals that many of the tested methods have similar performance when optimized, we find that the method based on a chi-squared optimization procedure is simplest to optimize, and has excellent temporal resolution. Finally, we apply these step detection methods to the question of observed step sizes for cargoes moved by multiple kinesin motors in vitro. We conclude there is strong evidence for sub-8-nm steps of the cargo's center of mass in our multiple motor records.
Key Words:
dynein, kinesin, molecular motors, multiple motors, myosin, step-size
