Search Results for Amir Lakizadeh

Abstract

Drug-drug interactions (DDIs) stand at the forefront of challenges in modern pharmacology, necessitating precise prediction methods to ensure patient safety. This study presents a pioneering approach that synergizes attributed heterogeneous graph embedding with deep learning to forecast DDIs and their specific classifications. Our methodology is delineated into two pivotal stages. The preliminary phase revolves around data assimilation, leading to the creation of specialized feature matrices such as Chemical Composition, Interaction Targets, Enzymatic Reactions, and Biological Pathways. These matrices culminate in a comprehensive drug network where drugs are symbolized as nodes. Upon rigorous data refinement, these matrices serve as attribute markers for each node. Capitalizing on the robustness of the attributed heterogeneous network, we amalgamate diverse drug attributes, thereby amplifying the depth of drug interaction assessments. The subsequent phase sees these drug embedding vectors undergo strategic concatenation, resulting in detailed feature vectors for drug pairings. The final step involves a dense neural network, tasked with decoding intricate drug interaction nuances. The introduction of an attention-driven embedding process further accentuates the model’s capability by emphasizing pivotal interactions. The promising results, coupled with an innovative methodology, sets the stage for future explorations, potentially revolutionizing DDI predictions.