Classification of diffraction patterns using a convolutional neural network in single-particle-imaging experiments performed at X-ray free-electron lasers

Abstract

Single particle imaging (SPI) at X-ray free-electron lasers is particularly well suited to determining the 3D structure of particles at room temperature. For a successful reconstruction, diffraction patterns originating from a single hit must be isolated from a large number of acquired patterns. It is proposed that this task could be formulated as an image-classification problem and solved using convolutional neural network (CNN) architectures. Two CNN configurations are developed: one that maximizes the F1 score and one that emphasizes high recall. The CNNs are also combined with expectation-maximization (EM) selection as well as size filtering. It is observed that the CNN selections have lower contrast in power spectral density functions relative to the EM selection used in previous work. However, the reconstruction of the CNN-based selections gives similar results. Introducing CNNs into SPI experiments allows the reconstruction pipeline to be streamlined, enables researchers to classify patterns on the fly, and, as a consequence, enables them to tightly control the duration of their experiments. Incorporating non-standard artificial-intelligence-based solutions into an existing SPI analysis workflow may be beneficial for the future development of SPI experiments.

The structure of tick-borne encephalitis virus determined at X-ray free-electron lasers. Simulations

“…In real SPI experiments, these steps cannot be avoided and have to be performed very carefully to obtain the final particle structure with high resolution. Each of the steps mentioned above have become the study of separate research: single hit classification (Bobkov et al, 2015;Shi et al, 2019;Cruz-Chu ´et al, 2021;Ignatenko et al, 2021;Assalauova et al, 2022), orientation determination (Loh & Elser, 2009;Ayyer et al, 2016) and background subtraction (Rose et al, 2018;Kurta et al, 2017;Lundholm et al, 2018).…”

“…The number of simulated diffraction patterns (1000) was based on the existing research (Poudyal et al, 2020) showing the dependence between this number, the experimental parameters and a spatial resolution of several nanometres or less. The number of simulated diffraction patterns can be also estimated from the experimental data (Rose et al, 2018;Assalauova et al, 2020Assalauova et al, , 2022. This number depends on the studied particle and experimental conditions.…”

“…For example, Assalauova et al (2020) used a clustering algorithm based on the maximum-likelihood method, which is actively used in cryo-EM (Dempster et al, 1977). In the work by Assalauova et al (2022), machine learning methods using a convolutional neural network (CNN) were used for the same purpose. The reconstructed electron density of the virus was obtained by the modal decomposition of several reconstructions of the virus under study (Assalauova et al, 2020).…”

Introduction to the virtual collection of papers on Artificial neural networks: applications in X-ray photon science and crystallography

“…Famous examples include recognizing handwritten digits and letters or identifying human faces. In photon science this is also a common use case, and recently we have, for example, seen it being used for femtosecond X-ray imaging patterns (FXI) (Assalauova et al, 2022), X-ray photon correlation spectroscopy (Timmermann et al, 2022) and serial femtosecond crystallography (Rahmani et al, 2023). This is often a convenient way to speed up a researcher's work by automating an otherwise labor-intensive task, by classifying a small set of patterns by hand and then training a machine-learning algorithm to classify a bigger dataset in a similar fashion.…”

Self-Supervised Approaches to the Classification of Spectra: Application to Phase Transitions in X-ray Diffraction Data

“…Methods based on machine learning (ML) are ideal for automation of repetitive tasks and identi cation of patterns in data sets, and several applications to data collected at x-ray facilities have been recently published (see, e.g., Ref. 8,9,10 ). When considering 1D spectral data, numerous classi cation approaches have been developed, including unsupervised clustering method such as spectral clustering 11 , K-Means 12 , Agglomerative clustering 13 , DBSCAN 14 , and supervised ML methods such as k-nearest neighbors 15 , partial least squares discriminant analysis 16,17 , decision trees 18 , random forests 19 , and extreme learning machines 20,21 .…”