CLigOpt: controllable ligand design through target-specific optimization.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Artificial intelligence in drug discovery

Background:

• Deep generative models face challenges in navigating vast molecular spaces for drug design.
• Fragment-based drug design (FBDD) offers a constrained approach to generate biologically active molecules.
• Optimization methods are crucial for identifying molecules with specific desired properties.

Purpose of the Study:

• To introduce CLigOpt, a controllable FBDD model for generating molecules with desired properties from fragment pairs.
• To leverage variational autoencoders and co-embeddings for comprehensive molecular graph information mining.
• To enable property-controlled molecular generation through a multi-objective Controllable Generation Module.

Main Methods:

• Utilized a variational autoencoder (VAE) architecture.
• Employed co-embeddings of node and edge features for molecular graph representation.
• Implemented a multi-objective Controllable Generation Module for property-guided generation.

Main Results:

• CLigOpt demonstrated strong performance in generating valid molecules across six metrics.
• Generated ligand candidates for hDHFR, increasing feasible active molecules by 10%.
• Applied molecular docking and synthesizability prediction to prioritize potential lead compounds.

Conclusions:

• CLigOpt effectively generates molecules with desired properties using a controllable FBDD approach.
• The model shows promise for efficient target-specific ligand design.
• Prioritization methods aid in deriving potential lead compounds from generated molecules.