A Deep Generative Model for Molecule Optimization via One Fragment Modification.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Molecule optimization is crucial for enhancing drug candidate properties via chemical modification.
• Existing methods face challenges in controlling optimization and maintaining molecular structure.

Purpose of the Study:

• To introduce Modof-pipe, a novel deep generative model for efficient molecule optimization.
• To demonstrate Modof-pipe's ability to retain molecular scaffolds and control optimization.

Main Methods:

• Developed Modof, a deep generative model operating on molecular graphs.
• Implemented Modof-pipe, a pipeline of Modof models for multi-site molecule modification.
• Enhanced Modof-pipe to Modof-pipe^\text{\textsuperscript{}} for generating multiple optimized molecules from one input.

Main Results:

• Modof-pipe successfully retains major molecular scaffolds and allows control over optimization steps.
• Achieved significant improvements in octanol-water partition coefficient, penalized by synthetic accessibility and ring size.
• Demonstrated superior performance over state-of-the-art methods, with up to 81.2% improvement without similarity constraints.

Conclusions:

• Modof-pipe offers a powerful and flexible approach for molecule optimization in drug development.
• The enhanced Modof-pipe^\text{\textsuperscript{}} further improves optimization efficiency and diversity.
• This model provides better control over molecular similarity and scaffold preservation during optimization.