Regeneration classifier audit: source-only reproduction package

Abstract

This package supports the study “Cohort-dependent normalization changes hierarchical assignments in a published regeneration classifier.” It contains analysis code, a Garnett compatibility adapter, saved perturbation definitions, comparison fixtures, 62 expected result tables, three figures, and reproducibility documentation for an audit using GSE205769. The analyses examine classifier compatibility, cell-type annotation, cohort-dependent normalization, gene masking, individual-gene sensitivity, and count thinning. This is an audit of a development-study dataset; it does not establish independent predictive accuracy or biological effects of gene perturbation. The package retrieves third-party expression data, the fitted classifier, and annotation references separately from recorded sources. These external inputs and installed software libraries are not bundled. Validation using freshly retrieved inputs completed all ten analysis stages. All 62 result tables matched the expected values, and all three regenerated figures matched. Testing used an existing compatible R installation; clean-machine dependency installation and cross-platform reproducibility have not been tested. Original code is licensed under MIT. Original documentation, figures, and results are licensed under CC BY 4.0, within the scope specified in LICENSE.md. Third-party materials retain their respective terms. This work received no external funding. All expenses were paid personally by the author.