Abstract
A new tool is introduced for screening macromolecular X-ray crystallography diffraction images produced at an X-ray free-electron laser light source. Based on a data-driven deep learning approach, the proposed tool executes a convolutional neural network to detect Bragg spots. Automatic image processing algorithms described can enable the classification of large data sets, acquired under realistic conditions consisting of noisy data with experimental artifacts. Outcomes are compared for different data regimes, including samples from multiple instruments and differing amounts of training data for neural network optimization.
Highlights
The recent introduction of X-ray free-electron laser (XFEL) light sources has made it possible to determine three-dimensional macromolecular structures from crystal diffraction patterns, acquired before radiation damage processes occur, that are generated from samples equilibrated at room temperature
As with traditional single-crystal diffraction protocols, it is important to arrange for data processing capabilities that produce real time feedback, in order to understand the characteristics of the experimental results
We envision the eventual application of convolutional neural network (CNN) as a screening tool in serial X-ray crystallography experiments, in which standard diffraction data are expert-annotated for when a CNN trained with one Rayonix dataset is applied to the other Rayonix datasets, where success rate is defined as the training process, after which the CNN can be deployed for real-time classification of new datasets from experiments number of correctly classified test images  100%: ð10Þ
Summary
The recent introduction of X-ray free-electron laser (XFEL) light sources has made it possible to determine three-dimensional macromolecular structures from crystal diffraction patterns, acquired before radiation damage processes occur, that are generated from samples equilibrated at room temperature. As data collection capacity continues to increase, it is important to assure that the data production rate does not exceed the network bandwidth for transmission to such facilities To this end, it is critical to develop the ability to veto certain diffraction events (such as those images that contain no Bragg spots) at the time and place of data collection, so that network and data processing resources are not overloaded by data with little value. If implemented on dedicated hardware such as an energy-efficient neuromorphic chip (Merolla et al, 2014; Esser et al, 2016), this type of screening procedure could in the future be coupled directly with the imaging detector as part of the data acquisition system
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
