Inferring system properties from thermodynamic fluctuations : a tool development approach

Abstract

Biological systems are far from equilibrium which require novel tools for unraveling their complex behavior. This thesis focuses on developing a toolbox in order to understand properties of living systems from thermodynamic fluctuations. In the first chapter, I discuss a fluorescence imaging platform which allows 3D information combined with non-invasive and photostable probes named single-walled carbon nanotubes. The second chapter discusses an image processing algorithm for analyzing the fluorescence images acquired with the proposed custom-built microscope. I demonstrate its robust image reconstruction capability under dense scenes of fluorescence images with its inherent parallel nature which allows implementation on GPUs. Finally, I develop a framework which predicts system properties from thermodynamic fluctuations in a data-driven manner. The proposed framework uses feature extraction methods based on wavelets with recurrent neural networks for processing time series data. A combination of these tools completes a pipeline which allows studying complex behavior of biological systems.