Restoring the tumor-suppressor function of p53 by inhibiting its negative regulator, MDM2, represents a significant therapeutic avenue for cancers that maintain wild-type p53. This research aimed to identify new MDM2 inhibitors through a phylogenetically guided strategy that involved the construction of a focused virtual library of metabolites derived from the Penicillium genus. A comprehensive computational framework was developed, employing machine learning-based quantitative structure-activity relationship (ML-QSAR) modeling, ensemble molecular docking, network pharmacology, molecular dynamics (MD) simulations, and ADMET profiling. The gradient boosting ML-QSAR model achieved a test set R2 of 0.80 and was externally validated against 39 known MDM2 inhibitors (R2 = 0.82, RMSE = 0.80 pIC50 units), confirming its predictive reliability. Ensemble docking studies against 13 conformations of MDM2 highlighted three leading candidates (CNP0147553.1, CNP0154476.3, and CNP0154476.4) demonstrating binding affinities comparable to the known control inhibitor Nutlin-3a, with docking scores validated against experimental binding data. Further investigations through 500 ns MD simulations provided insights into the stability of the CNP0147553.1-MDM2 complex, which maintained a mean ligand RMSD of 0.039 nm and a complex RMSD of 0.176 nm, alongside a favorable binding free energy of -25.82 kcal/mol. Key residue analysis revealed that CNP0147553.1 achieved pronounced stabilization of critical binding pocket residues, including an 81.6% reduction in flexibility of HIS96. Network pharmacology analysis revealed a polypharmacology potential, indicating that the hub genes related to the identified compounds predominantly converged on the PI3K-AKT-mTOR and RAS-RAF-MAPK signaling pathways. ADMET profiling suggested promising pharmacokinetic and safety profiles for the lead candidates, establishing the basis for future experimental validation.