Serial electron crystallography: merging diffraction data through rank aggregation

Abstract

Serial electron crystallography is being developed as an alternative way to collect diffraction data on beam-sensitive polycrystalline materials. Merging serial diffraction data from a large number of snapshots is difficult, and the dynamical nature of electron diffraction prevents the use of existing methods that rely on precise measurement of kinematical reflection intensities. To overcome this problem, an alternative method that uses rank aggregation to combine the rankings of relative reflection intensities from a large number of snapshots has been developed. The method does not attempt to accurately model the diffraction intensity, but instead optimizes the most likely ranking of reflections. As a consequence, the problem of scaling individual snapshots is avoided entirely, and requirements for the data quality and precision are low. The method works best when reflections can be fully measured, but the benefit over measuring partial intensities is small. Since there were no experimental data available for testing rank-based merging, the validity of the approach was assessed through a series of simulated serial electron diffraction datasets with different numbers of frames and varying degrees of errors. Several programs have been used to show that these rank-merged simulated data are good enough for ab initio structure determination using several direct methods programs.