An advanced workflow for single particle imaging with the limited data at an X-ray free-electron laser

💡 Research Summary

This paper presents a comprehensive workflow that enables high‑resolution single‑particle imaging (SPI) with severely limited data acquired at the Linac Coherent Light Source (LCLS). The authors studied the bacteriophage PR772 using the AMO instrument, delivering 1.7 keV (0.729 nm) X‑ray pulses of ~4 mJ energy at a 120 Hz repetition rate. A pnCCD detector recorded diffraction patterns, but only one of its two panels was functional, reducing the usable detector area to 512 × 1024 pixels. From an initial 12 million frames, 190 000 hits were identified with the psocake/psana pipeline. Subsequent pre‑processing steps—pixel‑wise background subtraction based on intensity histograms, precise beam‑center refinement, and particle‑size filtering using power‑spectral‑density (PSD) fits to solid‑sphere models—reduced the dataset to 18 000 candidate patterns consistent with the expected 70 nm virus size.

The central methodological advance is the application of an expectation‑maximization (EM) algorithm, originally developed for unsupervised clustering in cryo‑EM, to classify single‑hit diffraction patterns. Unlike the angular X‑ray cross‑correlation analysis (AXCCA) used in earlier SPI studies, EM can handle the missing half‑detector data and the very low signal‑to‑noise ratio. The algorithm iteratively assigns probabilities for each pattern to belong to a set of randomly initialized clusters, allowing in‑plane rotations to accommodate random particle orientations. Five independent EM runs were performed; the intersection of the resulting cluster selections yielded a robust set of 1 085 high‑contrast single‑hit patterns. This EM‑based selection showed markedly higher PSD contrast and clearer fringe visibility than a manual selection of 1 393 patterns, confirming the superiority of the statistical clustering approach.

Orientation determination and three‑dimensional reconstruction were carried out with the Expand‑Maximize‑Compress (EMC) algorithm as implemented in the Dragonfly software suite. EMC combines the 2D diffraction patterns into a 3D intensity distribution, which was subsequently subjected to mode decomposition to extract the electron density without imposing symmetry constraints. The final reconstruction reached a spatial resolution of 6.9 nm (2π/q_max, with q_max ≈ 1 nm⁻¹), limited primarily by the low scattering intensity and the modest number of usable single hits. The authors note that previous work on the same virus (Rose et al., 2018) employed roughly ten times more single‑hit patterns, but the present workflow demonstrates that comparable structural information can be obtained even when detector failures and limited hit rates reduce the data volume dramatically.

The complete pipeline can be summarized as follows: (1) hit finding, (2) background and beam‑center correction, (3) PSD‑based particle‑size filtering, (4) EM‑based unsupervised clustering for single‑hit classification, (5) EMC orientation recovery and 3D intensity reconstruction, and (6) mode decomposition for final density map and resolution assessment. Each stage is designed to minimize data loss and maximize signal fidelity, making the workflow robust against common SPI challenges such as detector defects, low photon counts, and heterogeneous sample delivery.

Beyond the specific PR772 case, the authors argue that this workflow is broadly applicable to other viral and non‑crystalline biological particles, especially when experimental constraints (e.g., detector downtime, limited beamtime, or low particle concentrations) preclude the collection of large, high‑quality datasets. Future improvements—higher XFEL fluence, detectors with greater dynamic range, and optimized aerosol injection—could further push the attainable resolution toward the atomic scale, fulfilling the long‑standing promise of XFEL‑based single‑particle imaging.