Accelerating discovery of bioactive ligands with pharmacophore-informed generative models

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Deep generative models advance drug discovery but often lack structural novelty.
• Limited structural novelty restricts inspiration for medicinal chemists.

Purpose of the Study:

• To develop TransPharmer, a generative model for de novo molecule generation with enhanced structural novelty.
• To integrate ligand-based pharmacophore fingerprints with a GPT-based framework.

Main Methods:

• Developed TransPharmer, a generative model combining pharmacophore fingerprints and a GPT framework.
• Utilized unconditioned distribution learning, de novo generation, and scaffold elaboration.
• Validated efficacy through case studies on dopamine receptor D2 (DRD2) and polo-like kinase 1 (PLK1).

Main Results:

• TransPharmer demonstrated excellence in distribution learning, de novo generation, and scaffold elaboration.
• An exploration mode facilitated scaffold hopping, yielding structurally distinct yet related compounds.
• Three of four synthesized PLK1 inhibitors showed submicromolar activity; IIP0943 achieved 5.1 nM potency with high PLK1 selectivity.

Conclusions:

• TransPharmer successfully generates structurally novel and bioactive ligands.
• The model's unique exploration mode can enhance scaffold hopping for drug discovery.
• TransPharmer represents a promising tool for identifying novel therapeutic agents.