IDENTIFICATION OF POTENT NATURAL PRODUCT LEAD COMPOUNDS AGAINST TARGET PROTEINS VIA HYBRID IN SILICO TECHNIQUES

Background: Natural products provide structurally diverse scaffolds for anticancer drug discovery, but single-stage virtual screening often overweights one imperfect scoring function and can produce unstable rankings.

Objective: We developed a hybrid, multi-objective in silico framework for prioritizing natural-product-like compounds against three therapeutically relevant nodes of the EGFR/PI3K/AKT signaling axis: epidermal growth factor receptor (EGFR), phosphatidylinositol-3-kinase alpha (PI3Kα), and AKT1.

Method: A simulated benchmark library emulating open natural-product chemical space was encoded using physicochemical descriptors and circular fingerprints, filtered by applicability-domain and ADMET rules, scored by nested-cross-validation machine learning, and integrated with structure-based docking to experimentally determined PDB structures 1M17, 4JPS, and 3O96. The highest-ranked candidates were evaluated with synthetic 100-ns molecular-dynamics stability traces and MM-GBSA-like binding-energy surrogates. Rank robustness was tested by bootstrap resampling and consensus-weight perturbation.

Results: XGBoost produced the strongest simulated ligand-based discrimination (ROC-AUC 0.904; F1 0.833). Consensus integration reduced target-specific rank volatility relative to docking alone and prioritized myricetin, quercetin, luteolin, and fisetin. Myricetin showed the most favorable three-target docking profile (-9.4, -9.1, and -8.7 kcal/mol) and the strongest simulated EGFR MM-GBSA energy (-49.8 ± 3.1 kcal/mol), while maintaining a stable complex RMSD near 0.20 nm. Bootstrap analysis yielded a 0.91 probability that myricetin remained within the top three.

Conclusion: The benchmark demonstrates that uncertainty-aware fusion of ligand-based learning, docking, ADMET filtering, and dynamic-stability evidence can produce more robust natural-product prioritization than any single computational score. Prospective biochemical and experimentally executed simulation validation is required before biological efficacy is inferred.