Toxicity atlas of antibody–drug conjugates — analytical dataset (FAERS 2012Q4–2026Q1)

Curated dictionaries, 2 × 2 contingency counts and complete disproportionality output supporting a payload- and target-informed toxicity atlas of antibody–drug conjugates (ADCs), derived from the US FDA Adverse Event Reporting System (FAERS), quarters 2012Q4 through 2026Q1. The deposit covers 13 ADCs meeting a prespecified threshold of at least 100 primary-suspect reports (62,079 reports across 11 target antigens and six payload classes), nine prespecified clinical toxicity clusters, drug-, payload- and target-level reporting odds ratios with 95% confidence intervals and proportional reporting ratios, eight prespecified pairwise ADC comparisons, a serious-only sensitivity analysis, outcome-reporting summaries, ranked clinical Preferred Terms, Benjamini–Hochberg false-discovery-rate correction for every analytical family, and two report-handling sensitivity analyses in which the case-version-selection rule and the nullified-case exclusion are relaxed one at a time. Included Python code recomputes every estimate, p value and q value directly from the deposited contingency counts using only the standard library. The DuckDB pipelines that produced the counts are published as documentation of every selection rule, join and formula. The raw FAERS quarterly files are not redistributed and remain freely available from the FDA. Reporting odds ratios quantify disproportionate reporting, not incidence, attributable risk or causality. Licensing: data files under CC BY 4.0, code files under the MIT Licence