Analysis of single particle imaging experiments at X-ray Free-Electron Lasers

Abstract

Single particle imaging (SPI) is a novel technique in X-ray science aimed at reconstructing the three-dimensional structure of nanoscale objects. Studying the inner structure of biological particles has become increasingly crucial, as evidenced by the pandemic of coronavirus disease (COVID-19) showing the necessity of scientific development in this field. The main advantage of this approach is that atomic structures can be resolved in their native environment without crystallization.

SPI experiments require using electromagnetic radiation with a sub-nanometer wavelength (such as X-rays) sufficient to resolve the object's internal structure. Because of the weak interaction of X-rays with matter, high coherence and photon flux are required to resolve the finest features in the object. Due to the extreme radiation dose, the biological particles are destroyed in the scattering process. To record a diffraction pattern corresponding to the undamaged structure, the X-ray pulse must have a duration shorter than the typical timescale of the destruction process. Therefore, high-brilliance synchrotron light sources could not be used due to insufficient coherent flux in a single pulse that is required for recording enough signal. The development of X-ray sources that have a high intensity and short pulse duration - X-ray free-electron lasers (XFELs) - overcome this challenge.

In the SPI method, many identical particles of the investigated system are injected into the X-ray beam providing diffraction images in random orientations. The three-dimensional structure of the object is obtained by applying complex algorithms to the collected diffraction patterns. The size of one such dataset could exceed terabytes; this motivates the development and implementation of elaborate data analysis techniques that help to save expensive XFEL time and speed up data processing.

The first two parts of this Thesis are based on the methodological development of the SPI data analysis workflow. The experimental data was collected from the virus PR772 at the Linac Coherent Light Source (LCLS) at SLAC, Stanford, USA in the frame of the SPI consortium. As a result of the developed methodology, which includes machine learning object classification, a three-dimensional virus structure with a resolution below 10 nanometers was reconstructed. The comparison of the result with the cryogenic microscopy studies showed similar features and an overall agreement between both techniques. Due to the complexity and cost of the SPI experiments, the preparation is a time- and effort-consuming process that requires high-level planning. The third part of this Thesis explores the optimization of set-up parameters through the simulation of the SPI experiment with tick-borne encephalitis virus. These simulations contributed to the success of an actual experiment performed at the European XFEL in Hamburg, Germany.