Polypharmacy—the simultaneous administration of several drugs—heightens the likelihood of harmful reactions caused by interactions between medications. With the rising occurrence of polypharmacy, accurately predicting these interactions has become crucial in drug development and healthcare. Because clinical trials struggle to identify uncommon side effects linked to multiple drug use, researchers are creating computational models to forecast such adverse outcomes. This paper presents PU-MLP, a Multi-Layer Perceptron-driven framework for predicting the side effects of drug combinations. The study applies state-of-the-art machine learning approaches to uncover associations between pharmaceuticals and their adverse reactions. The framework operates in three main phases: first, it generates high-quality representations for individual drugs by combining a random forest classifier, Graph Neural Networks (GNNs), and dimensionality reduction techniques. Next, it uses Positive Unlabeled learning to handle data uncertainty. Finally, a Multi-Layer Perceptron forecasts polypharmacy side effects. Evaluation via 5-fold cross-validation demonstrates that the proposed method outperforms existing techniques, achieving AUPR, AUC, and F1 scores of 0.99, 0.99, and 0.98, respectively.
