Prospective de novo drug design with deep interactome learning

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• De novo drug design aims to create novel molecules with specific properties.
• Existing methods often require extensive task-specific training or fine-tuning.

Purpose of the Study:

• To present a "zero-shot" computational approach for de novo drug design using interactome-based deep learning.
• To generate drug-like molecules with tailored bioactivity, synthesizability, and novelty.
• To evaluate the framework for protein structure-based drug design by targeting human peroxisome proliferator-activated receptor gamma (PPARγ).

Main Methods:

• Utilized a deep learning framework combining graph neural networks and chemical language models.
• Employed interactome-based learning for ligand- and structure-based molecule generation.
• Generated potential ligands for PPARγ, synthesized top candidates, and performed multi-faceted characterization.

Main Results:

• Successfully generated novel, drug-like molecules with specific bioactivity profiles.
• Identified potent PPARγ partial agonists with favorable selectivity.
• Confirmed the binding mode through crystal structure determination of the ligand-receptor complex.

Conclusions:

• The interactome-based de novo design framework enables efficient generation of innovative bioactive molecules.
• This approach offers a powerful alternative to traditional drug design methods.
• Demonstrated success in identifying novel ligands for a key therapeutic target (PPARγ).