Selecting XFEL single-particle snapshots by geometric machine learning.

Area of Science:

• Structural Biology
• Biophysics
• Data Science

Background:

• Single-particle imaging using X-ray Free-Electron Lasers (XFELs) offers a crystallization-free method for examining biological samples at room temperature.
• Obtaining high-resolution structures necessitates analyzing a large number of diffraction patterns from individual molecules, termed single particles.
• Efficiently identifying these single particles within massive XFEL datasets, especially at low signal levels and in the presence of background noise, remains a challenge.

Purpose of the Study:

• To present an efficient method for identifying single particles in large XFEL datasets.
• To develop a technique that is robust to low signal levels and background noise.
• To demonstrate the method's effectiveness on both simulated and experimental XFEL data.

Main Methods:

• Utilized supervised Geometric Machine Learning (GML) for feature extraction from training datasets.
• Fused test datasets into the established feature space for classification.
• Implemented a binary classification approach to distinguish 'single particles' from 'non-single particles'.

Main Results:

• Achieved near-perfect separation of single particles on noise-free simulated data.
• Quantified the method's predictive limits using simulated datasets with varying photon counts and background noise.
• Demonstrated superior performance on an experimental XFEL dataset compared to existing methods, covering a broad photon-count range.

Conclusions:

• Geometric Machine Learning provides an efficient and robust solution for single-particle identification in XFEL data.
• The GML method excels in identifying single particles even with significant background noise and low signal levels.
• This approach significantly advances the capability to retrieve structural information from challenging XFEL datasets, including those with structural variability.