TomoPicker: Annotation-Efficient Particle Picking in cryo-electron Tomograms.

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Accurate particle picking in cryo-electron tomograms (cryo-ET) is essential for in situ structural determination of macromolecules.
• Conventional template-matching methods are limited by template bias and low throughput.
• Learning-based methods face challenges due to the scarcity of annotated cryo-ET data, making manual annotation laborious.

Purpose of the Study:

• To develop an annotation-efficient particle picking method for cryo-ET.
• To address the data scarcity issue in training machine learning models for particle picking.
• To improve the throughput and accuracy of particle identification in cryo-ET datasets.

Main Methods:

• TomoPicker treats particle picking as a voxel classification task.
• The method employs two distinct positive-unlabeled learning strategies.
• It requires only a small fraction (~0.3-0.5%) of annotated particles for training.

Main Results:

• TomoPicker demonstrated high performance on a benchmark eukaryotic cell cryo-ET dataset.
• The method achieved approximately 30% improvement compared to state-of-the-art annotation-efficient approaches.
• Successful particle picking was achieved with minimal annotated data.

Conclusions:

• TomoPicker offers a robust and efficient solution for particle picking in cryo-ET.
• The annotation-efficient nature of TomoPicker significantly reduces the burden of data preparation.
• This method advances in situ structural biology by enabling more accessible and accurate macromolecular complex analysis.