Informative alignment patterns between chemical structure and clinical signals support context-aware toxicity assessment

Drug safety remains a major challenge in drug development. In clinical practice, safety outcomes are influenced not only by chemical structure but also by patient characteristics and treatment conditions, limiting the ability of structure-based predictions to fully capture real-world adverse events. Here, we integrated post-marketing safety data from the FDA Adverse Event Reporting System (FAERS; 2004Q1–2025Q2) with chemical structure and drug label information. Using 875 drugs, we quantified concordance between structure-based similarity and FAERS safety signals, as well as agreement between FAERS signals and label-reported risks. To support large-scale, reproducible signal detection, we developed a dedicated analytical application. Based on these measures, drugs were classified into four safety patterns. Pattern 1 (56.4%) showed consistent alignment among chemical structure, real-world safety signals, and labels, indicating informative structure–safety relationships. Pattern 2 (38.4%) diverged from structural similarity yet agreed with label information, suggesting safety profiles influenced by individual biology or clinical context. Pattern 3 (2.7%) reflected treatment-dependent attenuation of expected risks, whereas Pattern 4 (2.5%) revealed unanticipated risks not predicted by structure or labels. Collectively, this framework provides a pattern-based perspective on drug safety and suggests that such stratification may facilitate drug repurposing and support structure–activity relationship analyses.