Integrating Terahertz Time-Domain Spectroscopy with microfluidic platforms and machine learning for protein hydration studies

Terahertz Time-Domain Spectroscopy (THz-TDS) has been widely adopted as a technique of choice for sub-millimetre research over the last three decades, thanks to its high signal-to-noise ratio (SNR), broad bandwidth and straightforward phase retrieval. It allows the probing of picosecond dynamics of biomolecules, transient processes in semiconductors, imaging for security and medical applications, among many others. By measuring the amplitude of the electric field, the phase of the spectral information is preserved, and so complex parameters are easily extracted. However, manual preprocessing of the data is still required and iterative, numerical fitting of the material parameters is often needed, resulting in complexity, loss of accuracy and inconsistencies between measurements. The difficulties associated with analysing samples exhibiting strong terahertz absorption only compound with the aforementioned challenges. Aqueous solutions of proteins and their the surrounding water, or protein hydration shell, are uniquely probable with terahertz radiation, if this attenuation of the water can be addressed. This thesis aims to tackle both of these challenges. The use of machine learning techniques for interpreting spectroscopic THz-TDS data is described, achieved by training artificial neural networks (ANNs) with large data sets of simulated light-matter interactions, resulting in a computationally efficient method for material parameter extraction. The trained model improves on the accuracy of analytical methods that need approximations while being easier to implement and faster to run than iterative root-finding methods. Furthermore, a terahertz compatible, surface tension confined polytetrafluoroethylene (PTFE) based microfluidic flow cell is presented. THz-TDS measurements of a range of concentrations of aqueous bovine serum albumin (BSA) are shown to demonstrate the device’s efficacy of probing protein hydration dynamics in a transmission configuration. The combination of the machine learning techniques and the microfluidic device comprised of such a chemically inert material presented here offer new analytical and practical techniques for the probing of notoriously difficult samples