Work

SQRCD counterfactual formula-discrimination benchmark: data, code and reproducibility materials

Year: 2026
Type: software
Abstract: V13 release notes Version-specific archive: https://doi.org/10.5281/zenodo.22103254 GitHub release: https://github.com/Jeerryyoung/SQRCD_EQF/releases/tag/v13.0.0 Added 15,000-replicate herb-count-only pseudo-formula benchmark for each of 15 formulae. 15,000-replicate target-degree-matched benchmark. 15,000-replicate joint target-plus-compound-degree benchmark. 15,000-replicate herb-label-permutation benchmark. Replicate-level target- and compound-degree balance diagnostics. Proportions of matching replicates with absolute standardized mean difference below 0.1. Conservative global Benjamini-Hochberg sensitivity across all 60 target-FDI tests. Cross-null exception-stability classification and convergence checks. Sequential Supplementary Information numbering with split target- and compound-degree balance tables. Interpretation Most observed target-FDI values were compatible with topology-controlled alternatives. F020 was the only false-discovery-rate-stable exception across all four null families. SQRCD showed a nominal joint-null deviation that did not survive correction across formulae. Breaking conceptual change The counterfactual benchmark is the primary contribution. The Evidence Qualification Framework remains a downstream claim-control layer.
Language: English
Subfield: Drug Discovery
Open Access status: green
OpenAlex ID: W7204451024
Cites:
Cited by:
Related to: 10