BASDet: Bayesian approach(es) for structure determination from single molecule X-ray diffraction images

Abstract

X-ray free electron lasers (XFEL) are expected to enable molecular structure determination in single molecule diffraction experiments. In this paper, we describe an implementation of two orthogonal Bayesian approaches, previously introduced in Walczak and Grubmüller (2014), capable of extracting structure information from sparse and noisy diffraction images obtained in these experiments. In the ‘Orientational Bayes’ approach, a ‘seed’ model is used to determine for every recorded diffraction image the underlying molecular orientation. The molecular transform of the irradiated molecule is obtained by aligning and averaging those images in three-dimensional reciprocal space. By contrast, in the ‘Structural Bayes’ approach, a real space structure model is optimized to fit best to an entire set of diffraction images. This approach is used in a Monte Carlo structure refinement procedure.Both presented approaches were implemented in C; previous tests (Walczak and Grubmüller, 2014) suggest that the algorithms are robust against low signal to noise ratios and can deliver high resolution structural information. Program summaryProgram title: BASDetCatalogue identifier: AEZH_v1_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEZH_v1_0.htmlProgram obtainable from: CPC Program Library, Queen’s University, Belfast, N. IrelandLicensing provisions: GPL version 3No. of lines in distributed program, including test data, etc.: 1881590No. of bytes in distributed program, including test data, etc.: 49039580Distribution format: tar.gzProgramming language: C (ANSI 99), Perl.Computer: Workstation (8 CPUs).Operating system: Linux.Classification: 3, 4.13, 16.1.External routines: GNU Scientific Library (GSL), Message Passing Interface (MPI) libraryNature of problem: Extracting structural information from sparse and noisy single molecule XFEL diffraction images.Solution method: Bayes’ formalism is used to calculate either molecular orientation probability distribution with the aim to align individual images; or, alternatively, to calculate directly structure probability given all collected images.Running time: The examples given:Orientation_Bayes—50 h on Ivy Bridge Cores Xeon E3-1270v2 (2×4×3, 5 GHz)Structural_Bayes—These take longer than Orientation_Bayes runs, but can be restarted from checkpoint files.