Scaffold-Constrained Molecular Generation.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Artificial Intelligence in Drug Discovery

Background:

• Lead optimization in drug discovery often requires molecules with specific structural scaffolds.
• Generative models for de novo drug design struggle to meet scaffold constraints, limiting their practical application.
• Existing methods lack efficient scaffold-constrained generation capabilities.

Purpose of the Study:

• To introduce SAMOA (Scaffold Constrained Molecular Generation), a novel algorithm for scaffold-constrained in silico molecular design.
• To enhance the practicality of generative models for de novo drug design by enforcing scaffold requirements.
• To enable the design of molecules optimized for specific properties within a defined chemical space.

Main Methods:

• Utilized a SMILES-based Recurrent Neural Network (RNN) generative model.
• Implemented a modified sampling procedure within the RNN framework for scaffold-constrained generation.
• Integrated reinforcement learning methods for property optimization and focused chemical space exploration.

Main Results:

• Successfully demonstrated scaffold-constrained generation capabilities across multiple tasks.
• Generated novel molecules around specific scaffolds from SureChEMBL chemical series.
• Designed novel active molecules for the Dopamine Receptor D2 (DRD2) target and predicted actives for the MMP-12 series.

Conclusions:

• SAMOA effectively addresses the challenge of scaffold-constrained molecular generation in drug discovery.
• The algorithm enhances the utility of generative models for de novo design and lead optimization.
• SAMOA facilitates the exploration of relevant chemical space for designing molecules with desired properties and scaffolds.