A convolutional neural network-based screening tool for X-ray serial crystallography.

Area of Science:

• Structural Biology
• Biophysics
• Data Science

Background:

• Macromolecular X-ray crystallography is crucial for determining protein structures.
• X-ray free-electron lasers (XFELs) generate high-intensity X-ray pulses for crystallography.
• Processing large volumes of XFEL diffraction data presents significant computational challenges.

Purpose of the Study:

• To introduce a new, automated tool for screening macromolecular X-ray crystallography diffraction images.
• To apply a deep learning approach for detecting Bragg spots in XFEL data.
• To evaluate the tool's performance on realistic, noisy datasets with experimental artifacts.

Main Methods:

• Development of a data-driven deep learning tool utilizing a convolutional neural network (CNN).
• Implementation of automatic image processing algorithms for classifying diffraction images.
• Comparison of CNN performance across various data regimes, including different instruments and training dataset sizes.

Main Results:

• The CNN effectively detects Bragg spots in macromolecular X-ray crystallography diffraction images.
• The automated tool demonstrates capability in classifying large datasets acquired under realistic conditions.
• Performance analysis shows robustness across different experimental setups and varying amounts of training data.

Conclusions:

• The developed deep learning tool offers an efficient method for screening XFEL diffraction data.
• Automated Bragg spot detection improves the analysis of large crystallographic datasets.
• The tool's adaptability to different data conditions highlights its potential for widespread application in structural biology.