ABCG2, also known as the Breast Cancer Resistance Protein (BCRP), is a multidrug efflux transporter that undermines the efficacy of many chemotherapies and photodynamic therapy by reducing intracellular drug accumulation. Its overexpression has been linked to poor clinical outcomes, making it an important target for inhibitor development. In this study, we combined state-of-the-art machine learning and physics-based methods to characterize ABCG2–ligand interactions at the residue level. Using the Boltz-2 diffusion-based co-folding model, AMBER molecular dynamics, and MM/GBSA free energy analysis, we generated and evaluated binding poses of clinically relevant inhibitors, including Ko143, lapatinib, and afatinib, across multiple ABCG2 structural states. Results revealed key residues critical for stabilizing inhibitor binding and highlighted discrepancies between computational affinity predictions and available experimental data. Complementary ICM-Molsoft docking provided independent validation of binding trends. Together, these analyses established a framework for integrating machine learning–guided docking with physics-based scoring to inform rational design of next-generation ABCG2 inhibitors. Future directions include longer timescale MD simulations, use of generative diffusion models, and experimental high-throughput screening to identify potent, clinically translatable inhibitors. This integrative approach offers a scalable strategy for overcoming ABCG2-mediated drug resistance in cancer therapy