Simulating the Cellular Context in Synthetic Datasets for Cryo-Electron Tomography.

Area of Science:

• Cellular and Molecular Imaging
• Computational Biology
• Biophysics

Background:

• Cryo-electron tomography (cryo-ET) offers high-resolution cellular visualization.
• Lack of realistic ground truth data hinders deep learning applications in cryo-ET.
• Existing simulators fail to capture complex low-order cellular features.

Purpose of the Study:

• To develop advanced models for simulating realistic cellular structures in cryo-ET.
• To generate diverse and representative synthetic datasets for training deep learning algorithms.
• To provide an open-source tool for creating cryo-ET synthetic data.

Main Methods:

• Implementation of geometric and organization models for macromolecules, membranes, and filamentous structures.
• Utilizing parametrizable stochastic models for diverse geometries and crowded environments.
• Development of a multiplatform open-source Python package for cryo-tomogram generation.

Main Results:

• Successful simulation of low-order cellular features, including macromolecular clusters and filamentous networks.
• Generation of high-diversity, representative datasets mimicking native cellular environments.
• Provision of distortion-free ground truth density maps alongside synthetic tomograms.

Conclusions:

• The developed models and open-source package enable the creation of realistic synthetic cryo-ET datasets.
• These datasets are effective for training generalizable deep learning algorithms.
• The tool facilitates advancements in cryo-ET data interpretation and cellular organization analysis.