Data for Evaluation design shapes estimates of cross-donor generalization in single-cell classification of cell-type

This deposit contains the datasets supporting the study “Evaluation design shapes estimates of cross-donor generalization in single-cell classification of cell-type” by Xingyi Chen, Jiabei Li, and Alexis Battle (2026). The study compares random cell-level splitting with donor-held-out evaluation for single-cell cell-type classification. The archive includes ten AnnData (.h5ad) files across five cohorts: PBMC, lung, kidney, pancreas, and Blood Atlas. Each cohort includes a preprocessing input object or intermediate checkpoint and a benchmark-ready dataset containing the final cell population, cell-type and donor annotations, highly variable gene expression, and precomputed representations where available. Blood Atlas does not include scVI embeddings because gene-level raw counts are unavailable. The included data/README.md documents file contents and usage. Extract data.zip into the repository root to restore the expected data/ directory structure. Code and reproduction instructions: https://github.com/danielchen05/scRNA-cross-donor-generalization Manuscript, dataset provenance, and original-study references: https://doi.org/10.64898/2026.10.26.761949