Date Published: March 01, 2017
Publisher: International Union of Crystallography
Author(s): Amit Sharma, Linda Johansson, Elin Dunevall, Weixiao Y. Wahlgren, Richard Neutze, Gergely Katona.
Distribution analysis of intensity observations in serial femtosecond crystallography data processing helps to separate Bragg reflections from the background detector response.
X-ray free-electron lasers (XFELs) are linear accelerator based X-ray sources that deliver a peak X-ray brilliance a billion times greater than synchrotron radiation (Emma et al., 2010 ▸). These revolutionary machines were foreseen to create new possibilities in life science (Neutze et al., 2000 ▸) and over the six years since XFELs have been available to users the rapid pace of development has been impressive (Schlichting, 2015 ▸). One major application of XFEL radiation is the development of serial femtosecond crystallography (SFX). A proof-of-principle study first performed at low resolution using crystals of photosystem I (Chapman et al., 2011 ▸) was shortly afterwards extended to high resolution (Boutet et al., 2012 ▸) and has since been applied to time-resolved X-ray diffraction (Tenboer et al., 2014 ▸), the study of protein–protein receptor complexes (Kang et al., 2015 ▸) and de novo phasing (Barends et al., 2014 ▸). Serial crystallography has also since been applied to studies using synchrotron radiation (Nogly et al., 2015 ▸) and is expected to become a broadly applied method at storage ring based microfocus beamlines.
SFX experiments can produce highly heterogeneous data and our primary purpose was raising the awareness of their incorrect treatment. We observed that in the SFX data the skewness of the intensity observations follows the characteristic trend of a protein Wilson plot; the same is not true for the intensity of the most frequent observations. The most frequent intensity observations do not appear to originate from Bragg reflections of protein. To deal with the weak diffraction images one possibility is to pre-filter the reflection observations using a dynamic resolution cutoff per image basis. This approach risks rejecting weak observations that otherwise contain useful information, and indeed we have not observed a sharp distinction between the diffuse scattering signal and Bragg reflections. Instead, the information content can be judged from the distribution of the entire data set and the described ex-Gaussian diagnostics and our intensity inference strategy is a step in this direction.