Disentangling analytical sensitivity and donor influence in single-nucleus depression data

This lightweight reproducibility archive accompanies “Disentangling analytical sensitivity and donor influence in single-nucleus depression data.” Version: MD3-20261103-V2-F03.Version-specific DOI: https://doi.org/10.5281/zenodo.23916675. This revised release separates the effects of filtering, normalization and variance modeling under matched donor counts and technical designs, and examines donor influence and conditional error–detection behavior. The study includes 51 source-overlapping comparisons from 71 donors, comparisons involving OLS, limma-voom and edgeR, 61,200 conditional benchmark records, 18,000 conditional planning records, and a descriptive catalogue of 68 genes showing internal influence stability. The source studies already used donor-level pseudobulk analyses. Additional analyses include correlation-aware pathway assessment, an independent paired-stimulation context, an expanded 46-donor major depressive disorder context with annotation-confidence sensitivity, accessibility mapping, and resources for future fixed-score study planning and selected-score robustness assessment. The F03 release refines documentation and packaging while retaining the F02 numerical computation freeze; it adds no donors, fitted models or simulation repetitions. The Light archive contains executable analysis and plotting code, result summaries, a 37-table workbook, the stability catalogue, selected primary parvalbumin-interneuron gene tables, source fingerprints, data dictionaries, provenance and validation records, and figures in PDF, SVG and PNG formats. Large full per-gene and donor-influence outputs, simulation-replicate files and TIFF figures are omitted from this Light variant. The omitted files are listed in documentation/FULL_ONLY_PAYLOAD.csv. Reproduction instructions and source-input requirements are provided in code/REPRODUCE.md and the source-input manifests. The validation records distinguish the primary computations from subsequent numerical checks. They document source-count reconstruction and numerical regeneration of specified outputs. Separate repeat executions of the benchmark and planning analyses covered five benchmark sentinel cases and the first 100 planning seeds, respectively; they were not complete second executions of all simulation runs. The 51 comparisons share donors. Discovery totals represent gene–comparison hypotheses, rather than unique genes or confirmed true positives. The 68-gene catalogue describes internal influence stability and does not establish independent biological replication. Conditional simulations do not estimate clinical major depressive disorder false-discovery rates or establish a universal ranking of analysis methods. Missing mappings for RNA age, post-mortem interval and tissue pH remain unresolved. The external broad-cell depression analysis does not constitute like-for-like validation of the female parvalbumin-interneuron findings. Third-party raw matrices, individual clinical tables, nucleus-level arrays, complete external gene-set files and installed software environments are not redistributed. Their sources, access requirements and available fingerprints are documented in the archive. Third-party materials remain subject to their providers’ terms. Project-authored derived results and documentation are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). Software code is distributed under the MIT License, as detailed in the included LICENSE documentation. Historical V1 DOI: https://doi.org/10.5281/zenodo.22956598. The historical version is retained for transparency; the revised analyses and their interpretation are described in this release. Correspondence: Cheng Zhu, School of Mental Health, Wenzhou Medical University, China.Email: [e-mail]: https://orcid.org/0009-0009-2146-5759