AI-Assisted Identification of a Putative Allosteric Ligand Targeting the CDK4/Cyclin D1 Protein-Protein Interface

Area of Science:

• Oncology
• Computational Chemistry
• Drug Discovery

Background:

• First-generation CDK4/6 inhibitors show clinical success but face acquired resistance.
• Targeting the ATP-binding pocket limits exploration of alternative regulatory sites.
• Protein-protein interactions (PPIs) present novel therapeutic targets in cancer.

Purpose of the Study:

• To identify a novel allosteric small-molecule inhibitor targeting the CDK4/Cyclin D1 PPI interface.
• To utilize the RapidFunnel-AI virtual screening pipeline for candidate identification.
• To explore alternative regulatory sites beyond the ATP-binding pocket of CDK4.

Main Methods:

• Employed RapidFunnel-AI, a virtual screening pipeline, starting with 50,000 ChEMBL molecules.
• Integrated topological scans, fragment-level electronic property enrichment, ADMET/PAINS filtering, and multi-stage docking (Vina-GPU, AutoDock-GPU).
• Utilized explicit-solvent molecular dynamics, contact-retention analysis, and MM-GBSA for candidate validation, prioritizing pose persistence and contact continuity over single scores.

Main Results:

• The pipeline narrowed the library to 43 topological candidates, 25 filtered ligands, and 9 for MD validation.
• Ligand_020 demonstrated a persistent binding mode at the CDK4/Cyclin D1 interface across multiple 500 ns simulations.
• In contrast, existing inhibitors (palbociclib, ribociclib) showed transient binding and dissociation at the same interface.

Conclusions:

• Single-score docking or MM-GBSA can yield false positives at shallow PPI interfaces.
• RapidFunnel-AI, integrating AI, MD, and energy analysis, successfully nominated Ligand_020 as a putative allosteric inhibitor.
• This approach provides a reusable platform for PPI-focused drug discovery in oncology.