Novel applications of generative adversarial networks (GANs) in the analysis of ultrafast electron diffraction (UED) images.

Inferring transient molecular structural dynamics from diffraction data is an ambiguous task that often requires different approximation methods. In this paper, we present an attempt to tackle this problem using machine learning. Although most recent applications of machine learning for the analysis of diffraction images apply only a single neural network to an experimental dataset and train it on the task of prediction, our approach utilizes an additional generator network trained on both synthetic and experimental data. Our network converts experimental data into idealized diffraction patterns from which information is extracted via a convolutional neural network trained on synthetic data only. We validate this approach on ultrafast electron diffraction data of bismuth samples undergoing thermalization upon excitation via 800 nm laser pulses. The network was able to predict transient temperatures with a deviation of less than 6% from analytically estimated values. Notably, this performance was achieved on a dataset of 408 images only. We believe that employing this network in experimental settings where high volumes of visual data are collected, such as beam lines, could provide insights into the structural dynamics of different samples.

Fixed-target serial oscillation crystallography at room temperature.

A fixed-target approach to high-throughput room-temperature serial synchrotron crystallography with oscillation is described. Patterned silicon chips with microwells provide high crystal-loading density with an extremely high hit rate. The microfocus, undulator-fed beamline at CHESS, which has compound refractive optics and a fast-framing detector, was built and optimized for this experiment. The high-throughput oscillation method described here collects 1-5° of data per crystal at room temperature with fast (10°â€…s-1) oscillation rates and translation times, giving a crystal-data collection rate of 2.5 Hz. Partial datasets collected by the oscillation method at a storage-ring source provide more complete data per crystal than still images, dramatically lowering the total number of crystals needed for a complete dataset suitable for structure solution and refinement - up to two orders of magnitude fewer being required. Thus, this method is particularly well suited to instances where crystal quantities are low. It is demonstrated, through comparison of first and last oscillation images of two systems, that dose and the effects of radiation damage can be minimized through fast rotation and low angular sweeps for each crystal.

Fixed target combined with spectral mapping: approaching 100% hit rates for serial crystallography.

The advent of ultrafast highly brilliant coherent X-ray free-electron laser sources has driven the development of novel structure-determination approaches for proteins, and promises visualization of protein dynamics on sub-picosecond timescales with full atomic resolution. Significant efforts are being applied to the development of sample-delivery systems that allow these unique sources to be most efficiently exploited for high-throughput serial femtosecond crystallography. Here, the next iteration of a fixed-target crystallography chip designed for rapid and reliable delivery of up to 11 259 protein crystals with high spatial precision is presented. An experimental scheme for predetermining the positions of crystals in the chip by means of in situ spectroscopy using a fiducial system for rapid, precise alignment and registration of the crystal positions is presented. This delivers unprecedented performance in serial crystallography experiments at room temperature under atmospheric pressure, giving a raw hit rate approaching 100% with an effective indexing rate of approximately 50%, increasing the efficiency of beam usage and allowing the method to be applied to systems where the number of crystals is limited.

A modular and compact portable mini-endstation for high-precision, high-speed fixed target serial crystallography at FEL and synchrotron sources.

The design and implementation of a compact and portable sample alignment system suitable for use at both synchrotron and free-electron laser (FEL) sources and its performance are described. The system provides the ability to quickly and reliably deliver large numbers of samples using the minimum amount of sample possible, through positioning of fixed target arrays into the X-ray beam. The combination of high-precision stages, high-quality sample viewing, a fast controller and a software layer overcome many of the challenges associated with sample alignment. A straightforward interface that minimizes setup and sample changeover time as well as simplifying communication with the stages during the experiment is also described, together with an intuitive naming convention for defining, tracking and locating sample positions. The setup allows the precise delivery of samples in predefined locations to a specific position in space and time, reliably and simply.