HODC-AES v1.0.0: A Reliability-Aware Benchmark for Higher-Order Drug-Combination Adverse-Event Signal Prediction

HODC-AES is a reliability-aware benchmark for higher-order drug-combination adverse-event signal prediction in computational pharmacovigilance.

This v1.0.0 release contains the full archival benchmark package, including processed benchmark artifacts, generated contrastive negatives, prediction outputs, paper-ready tables and figures, documentation, release metadata, and SHA256 checksums.

The benchmark is built from public FAERS quarterly reports and is designed for evaluation under random and temporal validation, hard-negative protocols, calibration analysis, top-k alert-budget evaluation, and temporal generalization-gap analysis.

Main release statistics:
- 973,898 positive signal candidates
- 1,500,000 generated contrastive negatives
- Five negative protocols: random, drug-count-matched, frequency-matched, one-drug-replacement, and hybrid-hard
- Zero known-positive overlap among generated negatives
- 165,878 unique drug combinations
- 12,311 unique reaction terms

Important: HODC-AES treats FAERS-derived labels as reported pharmacovigilance signal candidates, not as confirmed causal clinical drug-drug interactions. The resource is intended for research use only and not for clinical decision-making.