PickYOLO: Fast deep learning particle detector for annotation of cryo electron tomograms

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Particle localization in cryo-electron tomography (cryoET) is a manual bottleneck for subtomogram averaging (STA).
• Automating this step is crucial for efficient high-resolution structure determination.

Purpose of the Study:

• Introduce PickYOLO, a deep learning framework for automated particle picking in cryoET.
• Evaluate PickYOLO's performance on diverse biological samples.

Main Methods:

• Developed PickYOLO, a universal particle detector based on the YOLO (You Only Look Once) object recognition system.
• Trained PickYOLO using center coordinates of representative particles.
• Tested PickYOLO on single particles, filamentous structures, and membrane-embedded particles.

Main Results:

• PickYOLO achieves high-yield and reliable particle detection.
• Detection rates range from 0.24-3.75 seconds per tomogram.
• Identified particle numbers comparable to manual picking by experts.
• Demonstrated effectiveness across various particle types.

Conclusions:

• PickYOLO significantly reduces time and manual effort in cryoET data analysis for STA.
• Enables faster and more efficient high-resolution structure determination using cryoET.