On the release of cppxfel for processing X-ray free-electron laser images

Abstract

As serial femtosecond crystallography expands towards a variety of delivery methods, including chip-based methods, and smaller collected data sets, the requirement to optimize the data analysis to produce maximum structure quality is becoming increasingly pressing. Here cppxfel, a software package primarily written in C++, which showcases several data analysis techniques, is released. This software package presently indexes images using DIALS (diffraction integration for advanced light sources) and performs an initial orientation matrix refinement, followed by post-refinement of individual images against a reference data set. Cppxfel is released with the hope that the unique and useful elements of this package can be repurposed for existing software packages. However, as released, it produces high-quality crystal structures and is therefore likely to be also useful to experienced users of X-ray free-electron laser (XFEL) software who wish to maximize the information extracted from a limited number of XFEL images.

Keywords: X-ray free-electron lasers; XFELS; computer programs; data analysis; serial femtosecond crystallography.

Figure 1

Plotting calculated partiality (red line) against estimate of partiality (yellow fill) by comparing against the reference data set, plotted against the Ewald sphere wavelength for the midpoint of each reflection. These are two models plotted vertically in four resolution shells to 1.6 Å resolution. The left and right models use two different bases: on the left, post-refinement has been carried out with a bandwidth of 0.18% and 0° mosaicity. On the right, the bandwidth was lowered to 0.07% and the mosaicity was increased to 0.03°. Both the bandwidth model and the mosaicity model were calculated with a super-Gaussian exponent of 1.5. The four panels on each side correspond to reflections at increasing resolution. The agreement is fairly close for both models at mid to high resolution, but the mosaicity model has a more erratic structure at low resolution. Overall, the correlation coefficient is reduced for the mosaicity model (89%) compared to the bandwidth model (96%), which suggests that a bandwidth model is preferred for the highly ordered CPV17 polyhedrin crystals from which these data were collected.

Figure 2

Workflow for executing indexing, initial orientation matrix refinement and post-refinement of individual images.

Figure 3

Simple integration box available in cppxfel, defined by the three parameters shown specifying the size of the boxes of foreground, neither and background flags.

Figure 4

Example of an Ewald sphere wavelength histogram produced during initial orientation matrix refinement, which is indicative of a successfully refined image.

Figure 5

Plotting calculated partiality (red line) against estimate of partiality (yellow fill) by comparing against the reference data set, plotted against the Ewald sphere wavelength for the midpoint of every reflection under 2.5 Å resolution. These plots represent incorrect parameter choices. The mean wavelength of the X-ray pulse was manually reconfigured to be off by 0.005 Å and the effect of this error is shown in the top panel. The crystal is rotated by the minimization method far off the true solution in order to compensate for the incorrect wavelength, to the point that the true wavelength peak is no longer recognizable. The middle panel had the reciprocal lattice point (rlp) size inflated by 100%, also leading to an incorrect crystal rotation to compensate by broadening the distribution of wavelengths. Finally, the bottom panel shows a set of good parameter choices which lead to a good refinement solution. This image come from a data set collected on CPV17 crystals.