Code and derived data for cardiometabolic liabilities and erectile dysfunction: overlap-corrected Mendelian randomization and corpus-cavernosum cell enrichment

Version 2.1.0 adds executable frozen-input model audits, donor-deletion sensitivity analyses, and software provenance to version 2.0.0. The release independently reproduces 21 primary univariable IVW results, compares 70 multivariable estimate rows across two specifications and two exposure models, reproduces 18 baseline cell-type/gene-set summaries, and reports 144 leave-one-donor-out tests. It includes hypothetical covariance stress tests, weighted-regression and fixed-unit invariance checks, frozen derived inputs, runtime records, a historical software evidence inventory, and file checksums.

The original multivariable specification and an additional Z-score-derived specification are retained side by side. Different exposure and outcome units alone do not invalidate a model; the SNP-specific transformation changes the specification and is not proof of corrected causal effects. Z/sqrt(N)-derived estimates are approximate sample-size-standardized associations, not odds ratios or calibrated disease-liability effects. The two diabetic tissue donors in the source atlas had type 1 diabetes; this tissue analysis does not validate a type 2 diabetes mechanism.

Numerical reproduction was verified from an extracted archive using the documented installed environment. Full-source MRlap, upstream LD clumping, single-cell preprocessing, integration, and scDRS scoring were not rerun in this audit. Historical software versions not supported by saved evidence remain explicitly unknown. Legacy pipeline code is retained separately for provenance and is not a newly validated portable end-to-end workflow.

Raw GWAS files, individual genotypes, and raw transcriptomic matrices are not redistributed. MIT licensing applies to author-written code and does not override third-party data terms. This release does not establish independent causal effects, mediation, treatment efficacy, or a validated therapeutic target.