A pharmacophore-guided deep learning approach for bioactive molecular generation.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Rational molecular design for novel bioactivity is challenging, particularly for understudied targets.
• Existing methods often lack flexibility and struggle with generating diverse, highly effective molecules.

Purpose of the Study:

• To develop a novel deep learning approach, Pharmacophore-Guided deep learning for bioactive Molecule Generation (PGMG), for accelerated drug discovery.
• To enhance the generation of bioactive molecules with desired properties, including high docking affinity and novelty.

Main Methods:

• PGMG utilizes a graph neural network for encoding chemical features and a transformer decoder for molecule generation.
• A latent variable is incorporated to manage the complex mapping between pharmacophores and molecules, improving diversity.
• The approach integrates pharmacophore guidance for flexible and targeted molecule design.

Main Results:

• PGMG demonstrated superior performance in generating molecules with strong docking affinities compared to existing methods.
• The generated molecules exhibited high scores for validity, uniqueness, and novelty.
• Case studies in ligand-based and structure-based de novo drug design validated PGMG's effectiveness.

Conclusions:

• PGMG offers a flexible and effective computational tool for accelerating the drug discovery process.
• The approach successfully addresses challenges in designing novel bioactive molecules for new and understudied targets.
• PGMG's ability to generate diverse and potent molecules makes it valuable for medicinal chemistry and drug development.