High-speed classification of coherent X-ray diffraction patterns on the K computer for high-resolution single...

Area of Science:

• Structural biology
• X-ray imaging
• Computational science

Background:

• Single-particle coherent X-ray diffraction imaging (CXDI) offers sub-nanometer resolution for biological supra-molecules.
• Analyzing millions of noisy diffraction patterns with unknown orientations is a major challenge.

Purpose of the Study:

• To develop a high-speed, scalable classification scheme for noisy X-ray diffraction patterns.
• To enable real-time feedback for optimizing CXDI experiments.

Main Methods:

• A high-speed, scalable classification scheme was developed for the K computer (10 PFLOPS).
• The scheme classifies approximately 1 × 10^6 noisy diffraction patterns.
• Classification was performed in parallel using 255 separate 1-hour jobs on a 385-node system.

Main Results:

• Successfully classified ~1 × 10^6 noisy X-ray diffraction patterns.
• Demonstrated the feasibility of real-time classification for experimental feedback.
• Validated the scheme using simulated diffraction patterns.

Conclusions:

• The developed high-speed classification scheme effectively handles large datasets of noisy diffraction patterns.
• This approach significantly enhances the signal-to-noise ratio, crucial for high-resolution structural determination.
• The system is designed for real-time integration with X-ray free-electron laser facilities like SACLA.