Scalable 3D Reconstruction From Single Particle X-Ray Diffraction Images Based on Online Machine Learning.

Area of Science:

• Structural biology
• Biophysics
• Computational imaging

Background:

• X-ray free-electron lasers (XFELs) provide powerful tools for studying biomolecular structure and dynamics.
• High-repetition-rate XFELs enable single particle imaging (X-ray SPI) under near-physiological conditions.
• Current X-ray SPI reconstruction algorithms struggle with the large datasets from advanced XFELs.

Purpose of the Study:

• To develop an efficient framework for reconstructing 3D macromolecular structures from large X-ray SPI datasets.
• To address the limitations of existing algorithms in handling massive data generated by modern XFELs.
• To enable real-time processing and reconstruction of X-ray diffraction data.

Main Methods:

• Introduction of X-RAI, an online reconstruction framework utilizing a convolutional encoder for pose estimation.
• Implementation of a physics-based decoder with an implicit neural representation for 3D reconstruction.
• End-to-end, self-supervised learning approach for processing large-scale X-ray SPI data.

Main Results:

• X-RAI demonstrates state-of-the-art performance on simulated and experimental datasets.
• The framework successfully processes large datasets with millions of diffraction images.
• Achieved high-quality 3D reconstruction from weakly scattering individual biomolecules.

Conclusions:

• X-RAI represents a significant advancement in handling large X-ray SPI datasets.
• The framework enables efficient and high-quality 3D structure determination of biomolecules.
• This work facilitates a paradigm shift towards real-time data acquisition and reconstruction in X-ray SPI.