Luoshu-Markov Geometric Framework: Fiedler Eigenvalue for Druggability Prediction, Allosteric Site Discovery, and Vina Discrepancy Correction (n=123)

We introduce the Markov-Fiedler spectral complement to the Lo-Shu druggability framework. For n=123 DUD-E proteins (87 druggable, 36 undruggable), the Fiedler eigenvalue lambda2 of the residue-contact graph Laplacian significantly distinguishes druggable from undruggable proteins (drug=0.057 vs undrug=0.090, Mann-Whitney p=0.0028). Combined model achieves 5-fold CV AUC=0.8393. Lo-Shu Markov allosteric score recovers validated allosteric proteins (cMyc-Max, CaM, HIF-1alpha, XIAP, PTPB) from structural geometry alone. CDK2 Vina discrepancy analysis explains BMS387032 gap (+3.13 kcal) via entropic N/C lobe boundary binding.