Date: 14 November 2018
Time: 12:45 - 13:30 (Lunch available from 12:35)
Room: RA 1501 (Ravelijn)
Speaker: Katharina Proksch (University of Göttingen).
Title: Statistical Inference for Molecules
Abstract:
Nowadays it is possible to image fluorescent probes on a sub-diffraction spatial resolution due to super-resolving scanning microscopes like STED or RESOLFT. This recent increase in resolution allows to visualize biochemical processes on a nanometer scale such that inference on the level of single molecules is now a reasonable goal in fluorescence microscopy. However, these new imaging techniques do not provide directly accessible information about the (spa-tial and numeral) distribution of molecules, but only about the total brightness distribution. In this talk, we discuss options to infer on the number and locations of molecules in a given sample based on a hybrid algorithm, which offers both the segmentation of an image as well as estimates of the local numbers of molecules, while it preserves uniform confidence about all statements. The proposed method allows for the construction of a novel, automatized statistical analysis tool for scanning microscopy via a molecular map, that is, a graphical presentation of locations as well as local numbers of molecules and corresponding uniform confidence statements. It is based on a sound statistical model, which connects both the local brightness and molecule distributions with the fact that a single molecule can emit only one photon at a time (antibunching). More precisely, our method is built on rigor-ous statistical convolution modeling of higher order photon coincidences and an approach on hot spot detection in heterogeneous data via multiscale scan statistics. We demonstrate the functionality of the molecular map by means of data examples from STED fluorescence microscopy. This is joint work with Jan Keller, Axel Munk and Frank Werner.