Scalable 3D reconstruction for X-ray single particle imaging with online machine learning.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• X-ray free-electron lasers (XFELs) provide unique insights into biomolecular structure and dynamics.
• High-repetition-rate XFELs enable single particle imaging (SPI) under near-physiological conditions, capturing transient molecular states.
• Current SPI reconstruction algorithms struggle with the massive datasets from modern XFELs due to slow, memory-intensive processing.

Purpose of the Study:

• To develop an efficient and scalable online reconstruction framework for X-ray single particle imaging.
• To address the computational bottlenecks hindering the analysis of large XFEL datasets.
• To enable high-quality 3D structure determination from millions of diffraction images.

Main Methods:

• Introduction of X-RAI (X-Ray single particle imaging with Amortized Inference), an online reconstruction framework.
• Utilizing a convolutional encoder for amortized pose estimation across large datasets.
• Employing a physics-based decoder with an implicit neural representation for end-to-end, self-supervised 3D reconstruction.

Main Results:

• X-RAI achieves state-of-the-art performance in simulations and experimental settings.
• Demonstrated unprecedented ability to process large datasets containing millions of diffraction images online.
• Achieved high-quality 3D macromolecular structure reconstruction.

Conclusions:

• X-RAI represents a paradigm shift in X-ray single particle imaging, enabling real-time reconstruction.
• The framework significantly accelerates the analysis of large XFEL datasets.
• Facilitates a deeper understanding of biomolecular structure and dynamics through efficient data processing.