Smart de novo Macromolecular Structure Modeling from Cryo-EM Maps

Area of Science:

• Structural biology
• Biophysics
• Computational biology

Background:

• X-ray crystallography has been the traditional method for determining macromolecular structures.
• Advancements in hardware and software have led to increased popularity of cryogenic electron microscopy (cryo-EM).
• The number of cryo-EM maps in databases like EMDataResource has grown significantly.

Purpose of the Study:

• To present automated, data-driven, and artificial intelligence approaches for macromolecular modeling.
• To highlight advancements in software for de novo macromolecular complex modeling.
• To discuss the development of deep learning-based methods for cryo-EM data analysis.

Main Methods:

• Automated map processing and feature extraction from cryo-EM data.
• Deep learning-based modeling building using tools like DeepTracer, now including DNA/RNA modeling.
• Protein identification directly from cryo-EM maps using DeepTracer-ID.

Main Results:

• Development of an automated pipeline for macromolecular modeling.
• Successful implementation of DNA/RNA modeling capabilities in DeepTracer.
• Creation of DeepTracer-ID for protein identification from cryo-EM maps.
• Accumulated experience in applying deep learning to macromolecule modeling.

Conclusions:

• Deep learning-based methods offer powerful solutions for automating labor-intensive macromolecular modeling.
• Advanced software tools are crucial for leveraging the growing volume of cryo-EM data.
• These advancements facilitate a deeper understanding of cellular machinery and impact pharmaceutical development, particularly for vaccines.