Sc2Mol

Sc2Mol generates scaffold-based small molecules for drug discovery by producing and decorating molecular scaffolds to explore chemical space and optimize drug-like structures.

Key Features:

• Scaffold-Based Generation: Focuses on molecular scaffolds and operates without requiring predefined scaffold patterns.
• Two-Step Process: A variational autoencoder generates scaffold structures and a transformer model decorates those scaffolds with functional groups.
• Random Molecule Generation and Optimization: Combines scaffold generation and decoration to enable both random molecule generation and targeted scaffold optimization.
• Empirical Validation: Evaluated on drug-like molecule datasets, demonstrating distribution learning and molecule optimization capabilities.
• Rule Learning: Learns transformation rules to convert coarse scaffolds into more elaborate drug candidates.

Scientific Applications:

• Pharmaceutical Research: Automates scaffold generation and decoration to explore chemical space and discover or optimize candidate molecules with desired properties.

Methodology:

Represents molecules as SMILES strings and uses a two-step generative framework in which a variational autoencoder generates scaffold SMILES and a transformer model decorates those scaffolds into drug-like molecules.