Organic anion transporting polypeptides (OATPs) are crucial for hepatic drug uptake, influencing drug efficacy and toxicity. Predicting OATP-mediated drug-drug interactions (DDIs) is challenging due to limited structural data and inconsistent experimental OATP inhibition data across studies. This study introduces Heterogeneous OATP-Ligand Interaction Graph Neural Network (HOLI-GNN), a novel computational approach that integrates molecular modeling with graph neural networks to enhance the prediction of OATP-mediated drug inhibition. By combining ligand molecular features with protein-ligand interaction data, HOLI-GNN outperforms traditional ligand-based methods. HOLI-GNN achieved median F1 and AUC scores of 0.78 and 0.90, respectively, compared to ECFP- and RDKit-based models built upon XGBoost (F1: 0.68 and 0.78, respectively; AUC: 0.70 and 0.75, respectively). Beyond improving inhibition prediction, we characterize protein residues involved in inhibitory versus non-inhibitory drug interactions, specifically highlighting residues T42, F224, I353, F356, and F386. We speculate that local position shifts in these hydrophobic packing residues, or the inhibition thereof, may be an important aspect of competitive inhibition mechanisms. Our model enhances the performance of OATP inhibitor prediction and, critically, offers interpretable interaction information to inform future mechanistic investigations.