Physics-Driven Selective Drug Design via Ising Spin Glass Evolution: A Deep Learning-Free Pipeline Applied to BCR-ABL and EGFR Oncogenic Targets

Targeted cancer therapy relies on designing molecules that selectively bind oncogenic proteins while sparing their healthy counterparts. Current computational approaches depend on deep learning architectures (AlphaFold, DiffDock, RoseTTAFold) requiring GPU clusters and millions of training structures. We present an alternative paradigm: a complete de novo drug design pipeline built entirely on classical statistical mechanics—specifically, Ising Spin Glass Hamiltonians coupled with Darwinian evolutionary optimization. Our method formulates selective drug-target binding as a multi-objective thermodynamic optimization problem, simultaneously maximizing oncogenic target affinity while minimizing healthy tissue binding. We validate the pipeline on three clinically relevant oncogenic targets: (1) BCR-ABL fusion kinase (Chronic Myeloid Leukemia), (2) expanded BCR-ABL binding domain, and (3) mutant EGFR (Non-Small Cell Lung Cancer). Across all three scenarios, the pipeline achieves 100% reproducibility (zero variance across 5 independent trials per target) and discovers peptide candidates with Selectivity Indices ranging from 3.0 to 7.0. Independent thermodynamic validation via 3D lattice Simulated Annealing confirms spontaneous structural collapse at the global energy optimum (E = -11, 11 stable hydrophobic contacts). The entire pipeline executes in under 1 second on commodity CPU hardware without any neural network components.