Data-driven in silico toxicity profiling of marketed drugs and retrospective analysis of post-marketing withdrawal risk.

Drug safety remains a central challenge in drug development and post-marketing surveillance, as the limited size and duration of pre-approval trials constrain detection of rare or delayed adverse effects, leaving patients exposed to residual risk after launch. Using ProTox 3.0 in silico toxicity predictions and statistical modeling, we analyzed an observational cohort of nearly 2,000 historically marketed small-molecule drugs and identified hepatotoxicity as a robust and consistent determinant of pharmaceutical post-marketing withdrawal probability after adjustment for therapeutic context. Other organ toxicities and structural descriptors of the drug molecules exhibited additional, indication-specific associations with withdrawal status, indicating that ex-ante toxicity liability is heterogeneous across therapeutic areas. While our analysis does not establish causality, the stable correlations observed across toxicity profiles, structural characteristics, and clinical indications suggest that a substantial fraction of withdrawal risk is statistically predictable. Embedding these quantitative risk signals into early-stage decisions and targeted safety evaluation strategies offers a data-driven approach to reduce late-stage attrition and post-marketing withdrawals, thereby contributing to safer and more efficient drug development.