Enhancing cryo-EM maps with 3D deep generative networks for assisting protein structure modeling.

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• Cryo-electron microscopy (cryo-EM) is increasingly used to determine biological macromolecular structures.
• Low resolution (3-6 Å) in cryo-EM maps often hinders accurate computational structure modeling.
• Improving cryo-EM map quality is crucial for advancing molecular structure determination.

Purpose of the Study:

• To develop a novel computational method to enhance cryo-EM maps for improved protein structure modeling.
• To address the challenge of low-resolution cryo-EM data that limits standard modeling tools.

Main Methods:

• Introduction of EM-GAN, a 3D generative adversarial network (GAN).
• Training the GAN on high- and low-resolution cryo-EM density maps to learn density patterns.
• Modification of input cryo-EM maps to improve their suitability for structure modeling.

Main Results:

• EM-GAN was extensively tested on 65 cryo-EM maps (3-6 Å resolution).
• The method demonstrated substantial improvements in protein structure modeling.
• Enhanced maps facilitated the use of popular protein structure modeling tools.

Conclusions:

• EM-GAN effectively enhances cryo-EM maps, particularly those with borderline resolution.
• The tool assists in overcoming resolution limitations for molecular modeling.
• This approach has the potential to accelerate structure-based drug discovery and biological research.