Detecting protein and DNA/RNA structures in cryo-EM maps of intermediate resolution using deep learning

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• Cryo-electron microscopy (cryo-EM) generates density maps of macromolecular structures.
• Near-atomic resolution maps are common, but intermediate/low-resolution maps present challenges for structure determination.
• Extracting detailed structural information from low-resolution cryo-EM maps remains difficult.

Purpose of the Study:

• To develop a computational method for identifying nucleic acids and protein secondary structures in low-resolution cryo-EM maps.
• To improve the accuracy and efficiency of structure modeling from cryo-EM data.
• To provide a tool for analyzing cryo-EM maps where traditional methods are less effective.

Main Methods:

• Development of Emap2sec+, a computational method utilizing a deep Residual convolutional neural network.
• Application of Emap2sec+ to cryo-EM density maps ranging from 5 to 10 Å resolution.
• Testing the method on both simulated and experimental cryo-EM datasets.

Main Results:

• Emap2sec+ accurately identifies DNA/RNA and protein secondary structures in low-resolution cryo-EM maps.
• The method assigns structural labels with associated probabilities at each voxel.
• Demonstrated stable and high assignment accuracy for nucleotides and improved performance for protein secondary structures compared to previous versions.

Conclusions:

• Emap2sec+ is an effective computational tool for analyzing low-resolution cryo-EM maps.
• The method facilitates structure modeling by providing voxel-wise structural information.
• Emap2sec+ enhances the utility of cryo-EM data, particularly when high-resolution data is unavailable.