TL;DR
This paper introduces "cryptotomography," a novel method to reconstruct the 3D structure of tiny particles from single, randomly oriented X-ray diffraction patterns. By extending existing algorithms to handle noisy data and unknown particle orientations, they've taken a significant step towards imaging single biomolecules in their native state.
What Problem Does This Paper Address?
Imaging very small biological molecules, like proteins and viruses, is incredibly challenging. Traditional methods often require crystallizing the sample, which can alter its natural structure. Coherent diffraction imaging (CDI) offers a way to image individual particles without crystallization, but it typically requires knowing the particle's orientation. In experiments using ultrafast X-ray Free Electron Lasers (XFELs), particles are often injected randomly and their orientations are unknown. Furthermore, the X-ray pulses can be noisy and the detectors can suffer from saturation and background scattering, making it difficult to extract useful structural information. This paper tackles the problem of reconstructing the 3D Fourier intensity distribution of such particles from these noisy, randomly oriented single-shot diffraction patterns, a technique they term "cryptotomography."
Methods
The core of this work is an extension of the Expansion-Maximization-Compression (EMC) framework, a computational algorithm used for image reconstruction. The authors adapted this framework to handle several key challenges:
- Random Orientations: The algorithm is designed to work even when the orientation of each particle is unknown. It iteratively refines a model of the particle's 3D intensity distribution and the orientations of the individual diffraction patterns.
- Noisy Data: The method accounts for fluctuations in X-ray pulse fluence and significant background scattering from experimental setup elements. They developed a sophisticated background subtraction technique by comparing each "hit" (a diffraction pattern likely from a particle) to nearby "non-hits" and selecting the best subtraction based on signal characteristics.
- Data Loss: The algorithm accommodates missing data due to detector saturation or the excision of problematic pixels. The redundant representation of the 3D intensity within the EMC framework helps mitigate the impact of such data loss.
The process involves iteratively expanding the 3D intensity model into multiple "tomograms" (representing different orientations), maximizing the likelihood of the data fitting these tomograms, and then compressing these tomograms back into a single 3D intensity model. This cycle is repeated until a stable reconstruction is achieved.
Datasets
The experiment used diffraction patterns from iron oxide nano-particles. These particles were chosen for their relative simplicity and availability, serving as a testbed for the cryptotomography technique. The paper does not introduce a new benchmark dataset but rather uses experimental data from a specific setup.
Key Insights and Ideas
- Cryptotomography is Viable: The most significant insight is the experimental demonstration that reconstructing 3D Fourier intensities from randomly oriented, noisy single-shot diffraction patterns is achievable. This opens doors for imaging samples that are difficult to orient or crystallize.
- EMC's Robustness: The paper highlights the adaptability of the EMC framework to handle real-world experimental challenges like fluctuating pulse fluence and background noise, which are common in XFEL experiments.
- Unmeasured Orientations are Not a Showstopper: The algorithm's ability to infer particle orientations alongside the 3D structure is a crucial breakthrough, simplifying experimental requirements.
- Probing Particle Shape: Even with simple particles, the reconstructed 3D intensities allowed for a quantitative comparison to ideal ellipsoidal particles, validating the accuracy of the reconstruction and providing insights into particle dimensions.
Key Results
The authors successfully reconstructed the 3D Fourier intensities of iron oxide nano-particles. They demonstrated that their method could recover reasonable particle dimensions (principal radii of approximately 25 nm and 100 nm) by comparing the reconstructed intensities to those of an ideal ellipsoidal particle. The reconstructed relative fluence distribution of the hits also provided insights into experimental fluctuations. The R-factor comparison indicated a good fit between the reconstructed intensities and an ideal ellipsoidal model.
Limitations and Future Work
- Resolution Limit: The reconstructed data has a pixellation effect, leading to a resolution limit of approximately 38.3 nm in the processed hits.
- Particle Complexity: The current work used relatively simple, solid-of-revolution particles. The authors suggest that future work will focus on applying cryptotomography to more complex, biologically relevant molecules.
- Phase Retrieval: While the paper reconstructs the 3D Fourier intensities, retrieving the real-space contrast (which requires phase information) is mentioned as a future step, noting that pixellation effects will be a factor.
- Statistical Ambiguities: The paper acknowledges that while EMC circumvents many ambiguities, there can still be statistical degeneracies in the reconstruction, particularly concerning fluence and intensity scaling. They addressed this by choosing the reconstruction with the slightly higher log-likelihood.
Summary updated