Reduction and segmentation of 4D data cubes for high-content Analysis FLIM-FRET

Abstract

An initial investigation seeks to elucidate the effect of the number and location of time gates on model parameter estimation accuracy. In silico experiments are used to obtain preliminary results which are then validated via in vitro and in vivo experimental data. It will be shown that 10 equally-spaced time gates allow the robust estimation of parameters of interest. Next, an optimal experimental design approach relying upon sensitivity analysis is used to better quantify parameter estimation using various numbers of time gates. Results are validated via in vivo experiments and compared to those obtained in the previous investigation. Finally, methods for estimation improvement and data reduction are developed and applied to a hyperspectral FLIM-FRET system. A global analysis algorithm enables improved estimation error margins while reducing the amount of data needed for accurate results. Furthermore, global analysis reduces the effect of experimental variations across a sample.

Fluorescence lifetime imaging (FLIM) is a powerful optical imaging technique. The lifetime of a fluorescent molecule is capable of revealing important information about its local environment. When combined with Forster resonance energy transfer (FRET), FLIM-FRET can be used to indirectly measure nanoscale interactions such as protein-protein interactions, protein-DNA interactions, or protein conformational changes. These events are identified by estimating parameters from the collected FLIM-FRET images using various model-based approaches. The quantitative accuracy of these approaches is typically dependent on factors such as signal-to-noise ratio as well as the number of temporal points (or time gates depending on the method) acquired when sampling the fluorescence decays. For complex applications such as hyperspectral or in vivo imaging, collecting many time points results in lengthy acquisition times which suppress the wide-spread implementation of FLIM-FRET techniques. Additionally, these complex applications typically result in larger amounts of data which can be cumbersome and difficult to process without significant computational resources. In this work, I present several methods that enable the reduction and segmentation of the data cubes resulting from FLIM-FRET analysis.