Unraveling information excited and corrupted by noise with applications to single molecule biophysics

Abstract

Single molecule studies enabled by recent nano-interrogation tools, offer unique insights not available to bulk studies constrained by averaging effects, and have revealed mechanisms of many fatal diseases including Alzheimer’s and Duchenne Muscular Dystrophy. However, gleaning insights at the molecular scale poses several challenges due to the extremely small dimensions and forces involved, the inherent stochastic behavior of the molecules, and the limitations of the instrumentation. Tools and algorithms are thus crucial for extracting information from and for building models using the single molecule data. In the first part of this thesis, we demonstrate one such self-learning change detection algorithm which can operate without any prior information about the locations or the magnitude statistics of the steps. We show how this algorithm with provable guarantees can be adapted for applications such as the single molecule force spectroscopy of muscle proteins. In the next part of the thesis, we present our findings on the nano-mechanical properties of the muscle protein utrophin, enabled by our tools that automate data analysis and provide rigor to the conclusions. In the final part of the thesis, we demonstrate another avenue for intelligent algorithms to push the boundaries of single molecule science: working with limited data and providing measures of confidence on results obtained from non-equilibrium single molecule experiments.