Personalized Drug Toxicity Risk Assessment Model Using Machine Learning

Adverse drug reactions (ADRs) pose a serious threat to patient safety, causing needless economic expenses to the healthcare system and creating unnecessary hospitalizations. The traditional toxicity tests do not take into account the personal. history- physiology, lifestyle and drug use. To address these shortcomings, we propose a personalized risk assessment approach built on Graph Neural Networks (GNNs). This model stratifies individuals into Normal, Low, or High risk categories of drug toxicity by examining demographic, physiological, and behavioral data. Unlike GNNs, which capture elaborate feature interdependencies, such as the duration of addiction, the functioning of various organs, and the frequency of substance use in a unified graph, traditional machine learning models compartmentalize features. Our model was trained on a synthetic behavioral-health dataset and achieved high accuracy alongside enhanced interpretability. We also designed a practical web-based interface for real-world application. This work demonstrates the transformative potential of GNNs in personalized ADRs and clinical decisionmaking.