AI-ACCELERATED DRUG DISCOVERY
Targeting challenging protein from integrin family
Addressing difficult-to-target binding pockets and discovering potential binders

Background

  • Heterodimeric receptor from integrin family involved in protein-protein interactions (PPIs).
  • Promising for immunological disorders, inflammation and cancers.
  • No binding sites for small molecules are known.
  • No potent small molecules binders exist.
  • Challenging target, often deemed undruggable by small molecules.

Methodology

  • MD simulations of the full-length dimer to produce an ensemble of protein conformations.
  • Receptor.AI pocket detection pipeline revealed 4 tentative binding pockets including one cryptic pocket.
  • Stock ~5M chemical space used.
  • Two screening tracks for each pocket (considering binding sites with cations):AI-based initial virtual screening followed by ArtiDock AI docking.
  • Direct ArtiDock AI docking of ~100k cluster centres.
  • ~500 hit candidates provided for experimental validation.
  • Protein ligand in vitro displacement assay used.
Project workflow

Results

  • 51 hit compounds obtained (~10% hit rate).
  • Potency range 10-100 μM in displacement assay.
  • Experimental validation included single concentration primary assessment and dose response measurements for initial hits.
  • Extremely diverse hits: 49 distinct Murcko assemblies out of 51 hits → diverse pool of scaffolds for subsequent optimization.
  • Four previously unknown prospective binding pockets for small molecules are identified using Receptor.AI pocket detection workflow.
  • Three of them are targeted by experimentally validated hits:
    • Pocket I: 14 hits
    • Pocket II:  30 hits
    • Pocket III: 7 hits
  • Multiple mechanism of action are likely to be triggered by the binders in different pockets → High flexibility and robustness in subsequent discovery stages.
Spatial representations of three binding pockets targeted by hit compounds