Classification and assessment of retrieved electron density maps in coherent X‐ray diffraction imaging using multivariate analysis. (1st January 2016)

• Record Type: Journal Article
• Authors: Sekiguchi, Yuki; Oroguchi, Tomotaka; Nakasako, Masayoshi
• Is Part Of: Journal of synchrotron radiation. Volume 23:Part 1(2016)
• Page Start: 312
• Page End: 323
• Publication Date: 2016-01-01
• Subjects: coherent X‐ray diffraction imaging -- X‐ray free‐electron laser -- structure analysis of non‐crystalline particles; Synchrotron radiation -- Periodicals; Free electron lasers -- Periodicals; 539.73505
• DOI: 10.1107/S1600577515018202
• Languages: English
• ISSNs: 0909-0495

Abstract

Abstract : Coherent X‐ray diffraction imaging (CXDI) is one of the techniques used to visualize structures of non‐crystalline particles of micrometer to submicrometer size from materials and biological science. In the structural analysis of CXDI, the electron density map of a sample particle can theoretically be reconstructed from a diffraction pattern by using phase‐retrieval (PR) algorithms. However, in practice, the reconstruction is difficult because diffraction patterns are affected by Poisson noise and miss data in small‐angle regions due to the beam stop and the saturation of detector pixels. In contrast to X‐ray protein crystallography, in which the phases of diffracted waves are experimentally estimated, phase retrieval in CXDI relies entirely on the computational procedure driven by the PR algorithms. Thus, objective criteria and methods to assess the accuracy of retrieved electron density maps are necessary in addition to conventional parameters monitoring the convergence of PR calculations. Here, a data analysis scheme, named ASURA, is proposed which selects the most probable electron density maps from a set of maps retrieved from 1000 different random seeds for a diffraction pattern. Each electron density map composed of J pixels is expressed as a point in a J ‐dimensional space. Principal component analysis is applied to describe characteristics in the distribution of the maps in the J ‐dimensional space. When the distribution is characterized by a small number of principal components, the distribution is classified using the k ‐means clustering method. The classified maps are evaluated by several parameters to assess the quality of the maps. Using the proposed scheme, structure analysis of a diffraction pattern from a non‐crystalline particle is conducted in two stages: estimation of the overall shape and determination of the fine structure inside the support shape. In each stage, the most accurate and probable density maps are objectively selected. The validity of the proposed scheme is examined by application to diffraction data that were obtained from an aggregate of metal particles and a biological specimen at the XFEL facility SACLA using custom‐made diffraction apparatus.