New Statistical Inference Methods for DEER Spectroscopy on Proteins

Abstract

Double Electron-Electron Resonance (DEER) spectroscopy is a pulse-EPR experiment that can provide sub-ångström resolution distance measurements of proteins and other biomacro- molecules in the distance range of about 2-16 nm. It is growing in importance as a tool for structural biology and, due to its robustness towards difficult sample conditions, complements existing experimental and theoretical techniques quite well. An outstanding problem in the field is the analysis and interpretation of the data: currently popular workflows invite user bias and there is little to no quantification of uncertainty. This is a significant barrier to more widespread adoption of the technique and may call into question published results. Transforming time-domain DEER data into distance distributions is an ill-conditioned, inverse problem. This means that there is no analytical solution and some form of regularization is necessary, which precludes straightforward fitting and error analysis. The work in this thesis introduces a Bayesian statistical framework for the processing and analysis of DEER data. The result is a set of numerical tools that allow for automated and statistically robust analysis.