CHiMP: deep-learning tools trained on protein crystallization micrographs to enable automation of experiments

Area of Science:

• Biophysics
• Structural Biology
• Drug Discovery

Background:

• Protein crystallization is crucial for structural biology and drug discovery.
• Automated analysis of crystallization experiments can accelerate research.
• Microscopy generates large datasets requiring efficient analysis.

Purpose of the Study:

• To develop deep-learning tools for automated analysis of protein crystallization experiments.
• To integrate these tools into synchrotron beamlines for real-time monitoring and data collection.
• To enhance fragment-based drug discovery screening platforms.

Main Methods:

• Development of three deep-learning tools: a classification network and two object detection/instance segmentation networks.
• Utilized transfer learning with pre-trained deep-learning networks.
• Applied tools to analyze micrographs from protein crystallization experiments at Diamond Light Source (DLS).

Main Results:

• The classification network categorizes experimental outcomes from images.
• Object detection and instance segmentation tools identify and mask crystals and droplets.
• Successful integration of tools into VMXi and XChem platforms at DLS.

Conclusions:

• CHiMP tools automate critical steps in protein crystallization analysis.
• These tools streamline monitoring and data collection at synchrotron facilities.
• Automated analysis facilitates efficient fragment-based drug discovery pipelines.