National Research Centre "Kurchatov Institute"

Abstract. About 1% of diffraction images produced in coherent X-ray diffraction imaging experiments originate from a single particle of interest and only those images are suitable for structure reconstruction. Other images contain contributions from multiple particles, water or some contaminant. Selection of single particle images is required. A new classification method that is based on cross-correlation analysis were developed. The method was successfully applied to the experimental data, that contain diffraction images of the PBCV-1 virus and T4 bacteriophage. In this article we present classification results for diffraction images of seven biological particles with different symmetry. The results confirm the applicability of the proposed method for correct classification of diffraction images corresponding to different molecules. We also studied influence of particle symmetry type and volume of learning dataset to classification quality.

Key words: coherent x-ray diffraction imaging, classification, x-ray cross-correlation analysis, support vector machine.