Convolutional networks for supervised mining of molecular patterns within cellular context

Area of Science:

• Structural biology
• Cell biology
• Biophysics

Background:

• Cryo-electron tomography (cryo-ET) provides high-resolution structural insights into cellular components.
• Accurate segmentation and localization of macromolecular complexes are crucial for understanding cellular architecture and function.

Purpose of the Study:

• To introduce DeePiCt, an open-source deep learning framework for supervised segmentation and macromolecular complex localization in cryo-ET.
• To benchmark DeePiCt's performance against state-of-the-art methods using annotated experimental data.

Main Methods:

• Development of DeePiCt, a deep learning framework utilizing supervised segmentation.
• Comprehensive annotation of 20 cryo-ET tomograms from Schizosaccharomyces pombe.
• Benchmarking DeePiCt against existing approaches on annotated datasets.

Main Results:

• DeePiCt demonstrates superior performance in identifying low-abundance and low-density macromolecular complexes.
• Analysis of cellular ribosome subpopulations and their association with mitochondria and endoplasmic reticulum.
• High-quality predictions achieved on unseen datasets from different species within minutes.

Conclusions:

• DeePiCt offers a powerful and efficient tool for macromolecular complex localization in cryo-ET.
• The framework facilitates the study of cellular composition and organization.
• Availability of annotated data and pre-trained networks accelerates research in structural cell biology.