Facilitating model reconstruction for single-particle scattering using small-angle X-ray scattering methods

Shufen Ma, Haiguang LiuJournal of Applied Crystallography 2016 · DOI

TL;DR

This paper introduces a computational pipeline that leverages Small-Angle X-ray Scattering (SAXS) analysis to enable real-time model reconstruction from single-particle X-ray Free-Electron Laser (XFEL) scattering data. By converting complex single-particle scattering patterns into a virtual SAXS profile, researchers can rapidly assess and retrieve plausible structural models, significantly accelerating the analysis workflow for non-crystalline samples.

What Problem Does This Paper Address?

Determining the structure of single molecules or non-crystalline particles using X-ray Free-Electron Lasers (XFELs) is a major goal in structural biology. XFELs provide intense, ultrashort pulses that can capture scattering data before radiation damage destroys the sample. However, a significant challenge is that the orientation of these tiny particles is uncontrolled. To reconstruct a 3D model, the orientation of each individual scattering pattern must be determined and the data merged in 3D reciprocal space. This process is computationally intensive and time-consuming, often delaying analysis until after the experiment is complete. This paper addresses the need for a faster, more immediate way to evaluate the structural information contained within these XFEL datasets, particularly for samples where precise orientation control is impossible.

Methods

The core of this work is a pipeline that transforms single-particle XFEL scattering data into a virtual SAXS profile. Here's how it works:

  1. Data Collection & Selection: Raw scattering patterns are collected at high repetition rates (e.g., 120 Hz at LCLS). "Hit-finding" programs (like Cheetah, OnDA) are used to identify patterns originating from actual particles.
  2. Scattering Center Optimization: The center of the scattering pattern is precisely determined by optimizing Friedel symmetry within the flat region of the Ewald sphere.
  3. Intensity Profile Extraction: Each selected scattering pattern is converted into a 1D "intensity profile" by integrating intensity along the radial direction (after converting to polar coordinates).
  4. Virtual SAXS Profile Generation: These individual intensity profiles are summed up. If the particle orientations are random, this cumulative sum approximates a SAXS profile, which represents the low-resolution structural information. The convergence of this virtual SAXS profile is monitored.
  5. Model Retrieval: The generated virtual SAXS profile is then fed into the sastbx.shapeup program (part of the SASTBX package). This program searches a database of known structures (compiled from sources like PISA, PDB, CATH, SCOP) to find models whose calculated SAXS profiles best match the experimental data. It uses a chi-squared minimization approach and a golden section search to optimize the model's radius and find the best fit.

The novelty lies in the conversion of complex, orientation-dependent single-particle scattering data into a simplified, orientation-independent SAXS profile, which can then be analyzed using established and efficient SAXS model retrieval techniques.

Datasets

The paper utilizes two main types of datasets for evaluation:

  • BIOISIS database and data from Grant et al. (2011): These are described as "experimental SAXS data" and are used to test the performance of sastbx.shapeup in retrieving models that match known structures.
  • Nanorice particles (iron oxide nanoparticles): Data was downloaded from the Coherent X-ray Imaging database (Maia, 2012). This dataset was used to demonstrate the pipeline's application to XFEL single-particle scattering data.
  • Polystyrene dumbbell particles: A publicly available dataset from Starodub et al. (2012) was used to further illustrate the method's applicability to more complex, non-spherical particles.

The paper does not introduce a new benchmark dataset but rather applies the developed pipeline to existing experimental data.

Key Insights and Ideas

  • SAXS as a Bridge: The most significant insight is that the low-resolution information captured in the SAXS regime is sufficient for rapid model assessment, even when derived from high-resolution single-particle XFEL data. This bypasses the need for full 3D reconstruction for initial model evaluation.
  • Real-time Feedback Potential: By converting complex scattering data into a SAXS profile, the analysis can be performed much faster. This enables "real-time feedback" during an experiment, allowing researchers to quickly assess data quality, particle behavior, and potential structural hypotheses.
  • Model Degeneracy and Multiple Models: The paper highlights that SAXS profiles, due to their low information content, can sometimes match multiple different structural models. The sastbx.shapeup program's ability to retrieve the top ten best-matching models is crucial for addressing this "model degeneracy" and providing a set of plausible candidates rather than a single, potentially incorrect, solution.
  • Utility for Complex Samples: The method is shown to be effective not only for simple particles but also for more complex assemblies like polystyrene dumbbells, demonstrating its broad applicability.

Key Results

  • Performance of sastbx.shapeup: On existing SAXS data, the sastbx.shapeup program achieved excellent agreement with reference models in 49 out of 52 test cases, with retrieval times averaging approximately 1 minute on a single CPU.
  • Nanorice Particle Analysis: For nanorice particles, the pipeline successfully converted single-particle scattering patterns into a virtual SAXS profile. The sastbx.shapeup program retrieved models that were in good agreement with the experimental data, estimating particle diameters of 135 nm and 68 nm.
  • Polystyrene Dumbbell Analysis: The pipeline also worked for polystyrene dumbbell particles. sastbx.shapeup retrieved multiple plausible models, with one model (model 3) showing good agreement with the expected dumbbell shape and experimental data. The retrieved radius (89 nm) was close to the expected value (91 nm).
  • Convergence Monitoring: Figures 3(a) and 3(b) demonstrate that the virtual SAXS profile converges as more scattering patterns are included, and the Pearson correlation can be used to monitor this convergence.

Limitations and Future Work

  • False-Positive Patterns: The presence of "false-positive" scattering patterns (from non-sample sources) can cause fluctuations in correlation curves. While the current method shows convergence even with these, an outlier rejection scheme could improve accuracy.
  • Orientation Preferences: The current virtual SAXS profile generation assumes random particle orientations. If particles exhibit orientation preferences, weighting schemes would be needed to ensure fair representation of all orientations.
  • Model Degeneracy: As noted, SAXS can lead to multiple models fitting the data. While retrieving multiple models helps, further refinement or complementary data might be needed for definitive structure determination.
  • Database Completeness: The accuracy of model retrieval depends on the diversity and completeness of the model database. Compiling larger, more diverse databases is suggested for improved accuracy, though this will increase search time.
  • Future Work: The authors suggest that the outlier rejection procedure could be integrated into the model refinement stage. They also mention that the described procedures and data sets are available online, facilitating further research. The potential for using orientation preferences to aid structure determination is also hinted at.

Summary updated

Abstract

X-ray free-electron lasers generate intense femtosecond X-ray pulses, so that high-resolution structure determination becomes feasible from noncrystalline samples, such as single particles or single molecules. At the moment, the orientation of sample particles cannot be precisely controlled, and consequently the unknown orientation needs to be recovered using computational algorithms. This delays the model reconstruction until all the scattering patterns have been re-oriented, which often entails a long elapse of time and until the completion of the experiment. The scattering patterns from single particles or multiple particles can be summed to form a virtual powder diffraction pattern, and the low-resolution region, corresponding to the small-angle X-ray scattering (SAXS) regime, can be analysed using existing SAXS methods. This work presents a pipeline that converts single-particle data sets into SAXS data, from which real-time model reconstruction is achieved using the model retrieval approach implemented in the software package <i>SASTBX</i> [Liu, Hexemer & Zwart (2012). <i>J. Appl. Cryst.</i><b>45</b>, 587-593]. To illustrate the applications, two case studies are presented with real experimental data sets collected at the Linac Coherent Light Source.
BibTeX
@article{ma2016facilitating,
  author = {Shufen Ma and Haiguang Liu},
  title = {{Facilitating model reconstruction for single-particle scattering using small-angle X-ray scattering methods}},
  journal = {Journal of Applied Crystallography},
  year = {2016},
  doi = {10.1107/s1600576716004337},
}
Figure