Classification of diffraction patterns using a convolutional neural network in single-particle-imaging experiments...

Area of Science:

• Structural Biology
• Biophysics
• X-ray Science

Background:

• Single particle imaging (SPI) at X-ray free-electron lasers (XFELs) is crucial for room-temperature 3D particle structure determination.
• Isolating single-hit diffraction patterns from vast datasets is a key challenge in SPI data processing.

Purpose of the Study:

• To investigate the application of convolutional neural networks (CNNs) for classifying diffraction patterns in SPI.
• To develop and evaluate CNN architectures for efficient and accurate selection of single-particle diffraction events.

Main Methods:

• Two CNN configurations were developed, optimizing for F1 score and recall.
• CNNs were integrated with expectation-maximization (EM) selection and size filtering.
• Performance was assessed by comparing reconstruction results and analyzing power spectral density functions.

Main Results:

• CNN-based pattern selection yielded comparable 3D reconstruction results to traditional EM selection.
• While CNN selections showed lower contrast in power spectral density, they enabled similar structural outcomes.
• The use of CNNs streamlined the SPI reconstruction pipeline and allowed for on-the-fly pattern classification.

Conclusions:

• CNNs offer a powerful artificial intelligence-based solution for enhancing SPI data analysis workflows.
• Implementing CNNs can lead to more efficient experimental control and potentially shorter experiment durations.
• This approach represents a beneficial integration of AI into SPI experiments for future advancements.