Advanced AI-Driven Prediction of Pregnancy-Related\nAdverse Drug Reactions

Ensuring\ndrug safety during pregnancy is critical due to the potential\nrisks to both the mother and fetus. However, the exclusion of pregnant\nwomen from clinical trials complicates the assessment of adverse drug\nreactions (ADRs) in this population. This study aimed to develop and\nvalidate risk prediction models for pregnancy-related ADRs of drugs\nusing advanced Machine Learning (ML) and Deep Learning (DL) techniques,\nleveraging real-world data from the FDA Adverse Event Reporting System.\nWe explored three methodsInformation Component, Reporting\nOdds Ratio, and 95% confidence interval of RORfor classifying\ndrugs into high-risk and low-risk categories. DL models, including\nDirected Message Passing Neural Networks (DMPNN), Graph Neural Networks,\nand Graph Convolutional Networks, were developed and compared to traditional\nML models like Random Forest, Support Vector Machines, and XGBoost.\nAmong these, the DMPNN model, which integrated molecular graph information\nand molecular descriptors, exhibited the highest predictive performance,\nparticularly at the preferred term level. The model was validated\nagainst external data sets from SIDER and DailyMed, demonstrating\nstrong generalizability. Additionally, the model was applied to assess\nthe risk of 22 oral hypoglycemic drugs, and potential substructure\nalerts for pregnancy-related ADRs were identified. These findings\nsuggest that the DMPNN model is a valuable tool for predicting ADRs\nin pregnant women, offering significant advancement in drug safety\nassessment and providing crucial insights for safer medication use\nduring pregnancy.