Cryo2StructData: A Large Labeled Cryo-EM Density Map Dataset for AI-based Modeling of Protein Structures

Area of Science:

• Structural Biology
• Biophysics
• Computational Biology

Background:

• Single-particle cryo-electron microscopy (cryo-EM) enables atomic-resolution structural determination of large biomolecules.
• High-resolution structures accelerate biomedical research and drug discovery.
• Automated atomic model building from cryo-EM maps is challenging, especially without templates, and current AI methods struggle with limited training data.

Purpose of the Study:

• To address the limitations of existing datasets for training AI models in atomic model building from cryo-EM density maps.
• To create a comprehensive and labeled dataset for developing and validating AI methods.

Main Methods:

• Development of Cryo2StructData, a dataset comprising 7,600 preprocessed cryo-EM density maps.
• Each voxel in the dataset is labeled with its corresponding atomic structure information.
• Training and testing of deep learning models using the Cryo2StructData dataset.

Main Results:

• Cryo2StructData is significantly larger than existing publicly available datasets for this purpose.
• Validation through training and testing deep learning models demonstrated the dataset's quality.
• The dataset is suitable for training and testing AI methods for atomic model building.

Conclusions:

• Cryo2StructData provides a robust foundation for advancing AI-driven atomic model construction in cryo-EM.
• This resource is expected to accelerate the process of determining molecular structures, benefiting structural biology and drug discovery.
• The availability of this large, labeled dataset will facilitate the development of more accurate and efficient AI tools.