A pocket-based 3D molecule generative model fueled by experimental electron density.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Structural Biology

Background:

• Accurate three-dimensional molecular structures are crucial for drug discovery.
• Generating novel drug-like molecules that fit protein targets remains a challenge.
• Electron density (ED) provides detailed structural information for molecular design.

Purpose of the Study:

• To develop and validate a novel computational method for generating drug-like molecules using experimental electron density (ED) data.
• To assess the model's ability to create diverse and chemically valid molecules with potential therapeutic applications.
• To provide an accessible online service for academic researchers in drug discovery.

Main Methods:

• Utilized a generative adversarial network (GAN) to generate ligand ED within a target protein pocket.
• Developed an ED interpretation module for de novo molecule generation.
• Tested the model on three diverse protein targets: a kinase, a protease, and a nuclear receptor.
• Evaluated generated molecules against a large dataset of known active compounds.

Main Results:

• The model successfully generated drug-like molecules with structural similarity to known active compounds.
• Novel molecules were created that exhibit similar binding modes to existing active compounds.
• Generated molecules demonstrated chemical validity and diversity within chemical space.
• The approach showed promise for library generation and fragment-based drug design.

Conclusions:

• Experimental electron density is a viable training dataset for generating novel drug-like molecules.
• The developed model is a powerful tool for accelerating drug discovery and high-throughput virtual screening.
• The online service facilitates access to this innovative technology for the research community.