Allosteric Inhibition of the Oncogenic JAK2 V617F Pseudokinase Domain via Polar Heterocycles: Thermodynamic Deconvolution and Cross-Domain Selectivity Profiles

Background: Constitutive hyperactivation driven by the somatic V617F mutation in the Janus kinase 2 (JAK2) pseudokinase (JH2) domain disrupts the vital autoinhibitory interface governing the adjacent catalytic (JH1) domain. Because standard therapeutics target the structurally conserved orthosteric ATP pocket of JH1, they frequently induce severe kinome-wide off-target toxicities. This study establishes a rigorous biophysical framework to define the thermodynamic and structural parameters required to achieve cross-domain selectivity (avoiding the catalytic JH1 domain in favor of the pseudokinase JH2 domain) coupled with mutation-specific targeting within the V617F substitution within the pseudokinase cleft. Methods: A computationally filtered drug-like heterocyclic library was screened using convolutional neural network-augmented molecular docking against wild-type JH1 and mutant JH2 crystal structures. Structural durability and dynamic persistence were quantified across independent triplicate 200 ns explicit-solvent molecular dynamics trajectories. Results: Molecular Mechanics Generalized Born Surface Area (MM-GBSA) free energy deconvolution identified Candidate 306 as a premier computationally predicted mutant-selective anchor, yielding a total binding free energy of −32.13 kcal/mol. This affinity is powered predominantly by strong van der Waals interactions (−41.11 kcal/mol) combined with a highly rigid, deep free energy basin. Dynamic cross-correlation matrix mapping verified that Candidate 306 mechanically synchronizes the P-loop, hinge, and DFG motifs to suppress local thermal fluctuations. Concurrently, the halogenated Candidate 253 predicted directional halogen-mediated contact networks within the mutated binding pocket, supported by the deep-learning scoring function’s parameterization of halogen-acceptor geometric orientation. Conclusions: Overcoming historical kinase database bias requires targeted polar optimization and shape-complementary packing to achieve stable protein deactivation, providing a robust predictive foundation for future experimental validation.