Scoring-Function-Agnostic and -Dependent Outcomes in BCL-2 G-Quadruplex Ensemble Docking: Polar Interactions, Not Molecular Flexibility, Determine Rank Divergence
Abstract
NMR ensemble docking rankings depend on both conformer selection and scoring function choice, yet no framework exists for classifying which outcomes are robust to this choice (scoring-functionagnostic) and which are sensitive to it (scoring-function-dependent), and the molecular basis of scoring function discordance for nucleic acid targets remains uncharacterized. We address both questions through pocket-driven ensemble docking of the human BCL-2 promoter G-quadruplex (PDB: 2F8U, 10 NMR models), cross-validated with AutoDock Vina (7,469 compounds screened, 114 ensemble-validated) and rDock (7,296 screened, 113 validated). Scoring-function-agnostic results include: the analytical framework (pocket-defined docking, Kendall's W concordance, sigma-scaling calibration), G4 vs. negative control discrimination direction, the W category hierarchy (G4-specific > overall), and the G4-specific cross-scoring rank correlation (Spearman rho = 0.632, p = 0.009, n = 16). Scoring-function-dependent results include: discrimination magnitude (Vina AUROC = 0.748, p = 0.010 vs. rDock AUROC = 0.602, p = 0.162), compound rankings (overall rho = 0.263, p = 0.111, n = 38; ChEMBL subset rho =-0.450, p = 0.027), and model concordance patterns. Molecular descriptor analysis revealed that hydrogen bond donor count (rho = 0.647) and topological polar surface area (rho = 0.525) predict rank divergence, while rotatable bond count does not, indicating that polar interaction parametrization-not conformational sampling-drives scoring function disagreement. Single-model screening recovered only 29-74% of the ensemble top-20, with Quarfloxin/CX-3543 as the only compound maintaining top-5 placement across both scoring functions. Ensemble docking methodology generalizes across scoring functions, while compound prioritization requires multi-scoring-function consensus, particularly for polar compounds.
Fields of Study & Subjects
Computational Drug Discovery Methods
DNA and Nucleic Acid Chemistry
Protein Structure and Dynamics
Publication details
- Journal / source
- ChemRxiv
- Publication date
- 17 February 2026
- Language
- EN
- Document type
- Preprint