Drug-drug interactions represent a critical challenge in the fields of medicine and pharmaceutical sciences, potentially resulting in severe adverse effects and mortality. The accurate prediction of these interactions, particularly for novel pharmaceutical compounds, remains a significant research challenge. This study presents a multimodal deep learning model for the prediction and classification of drug-drug interactions, engineered through the sophisticated integration of three distinct pharmaceutical data modalities: molecular structure, drug similarity metrics, and textual information. The proposed architecture employs a graph neural network for molecular structure processing, implements the Jaccard similarity coefficient for computing drug similarities, and utilizes the SciBERT model for textual information extraction. Model eva‎luation was conducted across three distinct scenarios of varying complexity. The results demonstrate that the proposed model achieves superior performance compared to existing methodologies, particularly in identifying interactions among established pharmaceutical compounds. The multimodal architecture facilitates generalization to novel compounds and exhibits potential for clinical implementation, both in supporting physician decision-making processes and in pharmaceutical development pipelines. Keywords: drug-drug interactions, deep learning, graph neural networks, natural language processing, data mining, multimodal learning, interaction prediction, pharmaceutical safety.

مهران صفاياني

عبدالرضا ميرزايي

جلال ذهبي