Robust multi-parameter classification-QSAR-based prioritization: Evaluating ranking stability under weight...

Alzheimer's disease remains a major therapeutic challenge, with β-secretase (BACE1) representing a key target for reducing amyloid-β production. Computational drug discovery workflows increasingly rely on multi-parameter prioritization integrating predictive modeling, structural analysis, and pharmacokinetic profiling. However, such approaches commonly depend on heuristic weighting schemes, and the robustness of resulting rankings under weight uncertainty remains poorly characterized. In this study, we developed a biology-informed multi-parameter prioritization framework integrating meta-ensemble classification-QSAR modeling, molecular docking, residue-level interaction analysis, ADMET profiling, and molecular dynamics simulations for BACE1 inhibitor discovery. Beyond compound ranking, the framework was designed to systematically evaluate the stability and reliability of prioritization outcomes. A curated dataset of 16,196 compounds was screened, yielding 153 predicted actives and 111 drug-like candidates. The meta-ensemble classification-QSAR model demonstrated strong predictive performance (accuracy = 0.852; ROC-AUC = 0.920), supported by cross-validation, external validation, and Y-randomization (p = 0.009). To address uncertainty in weight assignment, ranking robustness was quantitatively assessed using global sensitivity analysis under controlled perturbations and randomized weighting schemes. Results showed that rankings remained highly stable under moderate weight perturbations (Spearman ρ ≈ 0.998 for ±10% and 0.963 for ±25%), with partial degradation under randomized weights (ρ ≈ 0.821), indicating that prioritization is primarily driven by integrated multi-parameter signals rather than specific weight configurations. Post hoc optimization further confirmed the consistency of prioritized compounds. While Mol-3 exhibited the most favorable predicted binding free energy, Mol-2 demonstrated the most balanced overall computational profile when binding stability, catalytic interaction persistence, ADMET characteristics, and multi-parameter ranking criteria were considered collectively. Unlike conventional multi-parameter QSAR and scoring approaches that rely on fixed weighting schemes, the framework explicitly quantifies ranking robustness through perturbation, ablation, and optimization analyses. The proposed framework offers a structured and interpretable strategy for robust computational compound prioritization and highlights the importance of robustness analysis in computational drug discovery workflows.