Multi-objective de novo drug design with conditional graph generative model.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Molecular modeling

Background:

• Deep generative models show promise for de novo molecule design.
• Existing graph generative models are often too general and computationally expensive.
• Prior work predominantly focused on generating SMILES strings rather than molecular graphs.

Purpose of the Study:

• To develop a novel de novo molecular design framework using sequential graph generators.
• To improve molecule generation efficiency and scalability for larger molecules.
• To address limitations of current graph-based and SMILES-based generative models.

Main Methods:

• Proposed a new framework based on sequential graph generators without atom-level recurrent units.
• Scaled the model to handle larger molecules from the ChEMBL database.
• Employed a conditional graph generative model for drug design tasks with multiple objectives.

Main Results:

• The graph-based model significantly outperforms SMILES-based models, particularly in the rate of valid outputs.
• The proposed method is more tuned for molecule generation and handles larger molecules effectively.
• Demonstrated successful application in generating compounds with specific scaffolds, drug-likeness, and synthetic accessibility.

Conclusions:

• The novel graph-based generative framework offers improved efficiency and validity for de novo molecule design.
• Conditional graph generation provides flexibility for multi-objective drug design tasks.
• The approach is effective for generating targeted compounds, including dual inhibitors.