A Hierarchical Free Energy Framework for Ranking CDK Inhibitors: From PMF Barriers to Alchemical Precision

Abstract

   Accurate prediction of binding free energies for structurally homologous kinase targets remains one of the most challenging tasks in computational drug discovery. CDK1 and CDK2 provide an exceptionally stringent benchmark: their active sites differ by only 0.72 Å backbone RMSD, yet experimental ITC data reveal up to 170-fold selectivity differences for certain inhibitors, corresponding to a ΔΔG of only 2.87 kcal/mol, which is smaller than the typical error of most computational methods. Here we establish and validate a three-tier hierarchical free energy protocol applied to five CDK–inhibitor complexes (Dinaciclib, AZD5438, and CGP74514A bound to CDK1 and CDK2). Tier 1 (MM/PBSA endpoint calculations) achieves a qualitative affinity ranking but fails quantitatively, yielding R ≈ 0.45 and a critical rank inversion for CDK2-Dinaciclib.