Conditional Molecule Generator

Conditional Molecule Generator generates molecular graphs for de novo drug design using conditional deep generative models.

Key Features:

• Graph-Based Generation: Employs sequential graph generators to create molecular graphs instead of SMILES strings and is optimized for molecule generation.
• Atom-Level Recurrent Units Avoidance: Operates without atom-level recurrent units, reducing the computational expense associated with those units.
• Deep Generative Models: Leverages advanced deep generative models for de novo molecule generation.
• Scalability: Scaled to accommodate significantly larger and more complex molecules from the ChEMBL database, expanding covered chemical space.
• Performance Superiority: Outperforms SMILES-based models by producing valid molecular outputs at a higher rate.
• Conditional Generation: Uses a conditional graph generative model to generate molecules according to multiple objectives and design criteria.

Scientific Applications:

• Scaffold-Based Generation: Generates compounds containing predefined molecular scaffolds to facilitate exploration of novel chemical entities.
• Drug-Likeness and Synthetic Accessibility: Produces molecules that meet specified drug-likeness criteria and synthetic accessibility considerations.
• Dual Inhibitor Design: Has been applied to generate dual inhibitors targeting JNK3 and GSK-3β.

Methodology:

Implements a conditional graph generative model using sequential graph generation techniques that focus on molecular graphs rather than SMILES and avoids atom-level recurrent units.