Direct Detection and Atomic Modeling of Ligands in Cryo-EM Maps Using Deep Learning

Area of Science:

• Structural biology
• Computational chemistry
• Drug discovery

Background:

• Cryogenic electron microscopy (cryo-EM) is vital for structure-based drug discovery, enabling the study of macromolecule-ligand interactions.
• Computational interpretation of ligand density in cryo-EM maps is challenging, especially with unknown ligand positions or low resolution.
• Current methods require high-resolution structures and known binding sites, limiting early-stage drug discovery.

Purpose of the Study:

• To develop an automated computational framework for ligand detection and atomic modeling directly from cryo-EM maps.
• To address limitations of existing methods in early-stage structure determination and low-resolution cryo-EM data.
• To provide a unified approach for ligand discovery and structural modeling in drug development.

Main Methods:

• Introduction of Emap2lig, a two-stage deep learning framework.
• Emap2lig-Find: Identifies ligand-associated densities in cryo-EM maps.
• Emap2lig-Build: Employs a diffusion-based generative model for atomic ligand structure reconstruction.

Main Results:

• Emap2lig enables automated ligand detection and atomic modeling from cryo-EM maps.
• Emap2lig-Find successfully detects ligand densities at resolutions as low as approximately 5 Å.
• The framework provides a unified approach for ligand discovery and modeling across various resolutions.

Conclusions:

• Emap2lig offers a significant advancement for structure-based drug discovery using cryo-EM data.
• The framework overcomes key challenges in ligand density interpretation and atomic modeling.
• Emap2lig facilitates more efficient and comprehensive ligand identification and characterization in drug development.