CryoDataBot: a pipeline to curate cryoEM datasets for AI-driven structural biology

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• Cryo-electron microscopy (cryoEM) enables atomic-resolution visualization of biomacromolecules.
• Artificial intelligence (AI) methods are crucial for automating atomic model building from cryoEM maps.
• High-quality, task-specific datasets are essential for AI-based modeling but are labor-intensive to create.

Purpose of the Study:

• To develop an automated pipeline, CryoDataBot, for generating robust datasets for AI-driven structural biology.
• To streamline data retrieval, preprocessing, and labeling with quality control and customization.
• To reduce manual workload and facilitate tailored dataset construction for specific AI model objectives.

Main Methods:

• CryoDataBot automates data retrieval, preprocessing, and labeling.
• The pipeline incorporates fine-grained quality control and flexible customization options.
• Effectiveness was validated by assessing training efficiency of U-Net models and retraining CryoREAD.

Main Results:

• CryoDataBot significantly improved training efficiency for U-Net models.
• The pipeline enabled rapid and effective retraining of the CryoREAD RNA modeling tool.
• Demonstrated efficient generation of robust datasets with customizable quality control.

Conclusions:

• CryoDataBot simplifies and automates the creation of high-quality datasets for AI in structural biology.
• The pipeline's flexibility supports diverse AI modeling applications and reduces manual effort.
• Enables researchers to easily tailor dataset construction for specific modeling needs, enhancing AI-driven structural biology research.