A peak-finding algorithm based on robust statistical analysis in serial crystallography

Abstract

The recent development of serial crystallography at synchrotron and X‐ray free‐electron laser (XFEL) sources is producing crystallographic datasets of ever increasing volume. The size of these datasets is such that fast and efficient analysis presents a range of challenges that have to be overcome to enable real‐time data analysis, which is essential for the effective management of XFEL experiments. Among the blocks which constitute the analysis pipeline, one major bottleneck is `peak finding', whose goal is to identify the Bragg peaks within (often) noisy diffraction patterns. Development of faster and more reliable peak‐finding algorithms will allow for efficient processing and storage of the incoming data, as well as the optimal use of diffraction data for structure determination. This paper addresses the problem of peak finding and, by extension, `hit finding' in crystallographic XFEL datasets, by exploiting recent developments in robust statistical analysis. The approach described here involves two basic steps: (1) the identification of pixels which contain potential peaks and (2) modeling of the local background in the vicinity of these potential peaks. The presented framework can be generalized to include both complex background models and alternative models for the Bragg peaks.

Data reduction for X-ray serial crystallography using machine learning

“…Techniques for detecting Bragg peaks in diffraction patterns are a crucial building block in both rejecting bad images and analysing data. Peak finding methods employ carefully crafted threshold mechanisms based on statistical frameworks to separate Bragg peaks from the background signal (Parkhurst et al, 2016;Hadian-Jazi et al, 2017. For example, Li & Zatsepin (2018) used a simple global threshold mechanism to separate the background signal from the Bragg peaks.…”

Heterogeneous Ice Growth in Micron-Sized Water Droplets Due to Spontaneous Freezing

“…There has been a wide range of segmentation techniques suggested previously to find regions of interest (ROI) in a 2D image [53][54][55][56]. There is, however, no single standard segmentation technique that can deliver adequate results in all types of objectives.…”

Pump-Probe Time-Resolved Serial Femtosecond Crystallography at X-Ray Free Electron Lasers

“…An SFX experiment produces millions of diffraction patterns comprising both hits and blanks (diffraction patterns without Bragg reflections). Separating hits from the pool of all detector images follows a well-established process and is achieved by software solutions such as CASS [61], Psocake [62,63], Cheetah [64] and others [65]. They also perform preprocessing steps such as background subtraction, electronic noise correction, bad and hot pixel masking, and flag Bragg spots in the diffraction pattern.…”