Donor-Aware Cell-Class-Resolved Mapping of Nutrition-Relevant Repair Programs in Healing and Non-Healing Diabetic Foot Ulcers
Description
This repository provides the complete current-state data, code, and audit resources supporting a donor-aware integrative transcriptomic analysis of nutrition-relevant repair programs in diabetic foot ulcer healing. The study combines two independently generated human single-cell RNA-sequencing datasets with bulk-transcriptomic contextual support and an exploratory GeoMx spatial-context assessment. The repository includes curated dataset metadata, processed-file provenance, frozen broad cell-class annotations, the prespecified eight-module NRC_primary_v1 gene-set system, gene-level source and evidence annotations, analysis-ready score tables, donor and biological-sample crosswalks, inferential outputs, figures, supplementary-table source data, R scripts, session-information records, MD5 manifests, and reproducibility audits. In GSE165816, 14 primary foot-skin expression matrices were reconstructed to 11 independent donors, comprising seven Healer and four Non-Healer donors. Repeated matrices from the same donor were pooled before cell-count eligibility assessment and inferential analysis. GSE231643 contributed eight independent biological samples, comprising five Healer and three Non-Healer samples. The current primary analysis contains 96 eligible dataset-specific contrasts, 32 cross-dataset-comparable cell-class–module combinations, 18 directionally concordant combinations, and eight donor-audited cross-dataset directional candidates. Three candidates showed interval- and stability-supported directional evidence, whereas five remained directionally consistent with broader uncertainty. No Benjamini–Hochberg-adjusted q value was below 0.05. A post hoc dataset-wise predicted-doublet sensitivity analysis assessed 72,262 cells and identified 3,677 predicted doublets (5.09%). Exclusion retained the direction of all eight primary candidates and the prespecified candidate classification of seven. This sensitivity analysis did not redefine the frozen primary eight-candidate set. This Version 2 release supersedes Version 1 for all current analyses, reproducibility assessment, secondary use, and manuscript reporting. Version 1 remains available only as an immutable historical snapshot. Its matrix/sample-level 9-Healer versus 5-Non-Healer inference, 104-contrast testing family, legacy 10-signal Tier A layer, and corresponding earlier Figure 5 and Supplementary Table S17 should not be used for current interpretation. Public source-expression files remain available through their respective GEO records. This repository contains derived analytical resources and does not introduce new participant-level clinical data.
Files
Steps to reproduce
1. Download and extract the complete Version 2 release archive. Version 2 supersedes Version 1 for current analysis and manuscript reporting; Version 1 is historical only. 2. Read 00_README_FIRST.txt, 00_VERSION_2_CORRECTION_AND_SUPERSESSION_NOTICE.txt, 00_EXECUTION_ORDER.csv, 00_FILE_GOVERNANCE_FINAL.csv, and 00_RELEASE_MANIFEST.csv to identify active outputs, superseded files, execution order, and MD5 checksums. 3. Obtain the processed files and metadata for GSE165816, GSE231643, GSE143735, GSE134431, GSE199939, and GSE166120 from GEO. Update local paths and verify filenames. The workflow used processed matrices rather than FASTQ-level reprocessing. 4. Reproduce the stable upstream workflow: metadata audit; scRNA-seq object construction; dataset-specific QC and filtering; broad cell-class annotation; NRC-ready preparation; freezing of the eight NRC_primary_v1 modules; and cell-level scoring. Matrix/sample summaries are intermediate audit records, not inferential replicates. 5. Apply the Step21A3 crosswalk to map 14 GSE165816 foot-skin matrices to 11 reconstructed donors (7 Healer, 4 Non-Healer). Pool repeated matrices within donor, then reapply the prespecified ≥30-cell rule. Retain the eight original GSE231643 samples (5 Healer, 3 Non-Healer) as inferential units. 6. Run Step21B using reconstructed donors for GSE165816 and original biological samples for GSE231643. Rebuild biological-unit × cell-class × module summaries, 96 eligible contrasts, bootstrap intervals, exact biological-unit-label permutation tests, BH adjustment across all 96 contrasts, and leave-one-biological-unit-out analyses. Cells and repeated matrices are not independent replicates. 7. Run Step21C and the final PATCH2 layer to reconstruct eight donor-audited cross-dataset directional candidates and 16 dataset-specific estimates. Do not restore the superseded 104-contrast family, 10-signal Tier A layer, old Figure 5, or old Supplementary Table S17. 8. Run Step21D to generate the current figures, tables, supplementary outputs, and replacement records. Treat bulk RNA-seq and GeoMx analyses as contextual support only, not validation of cell-class-specific scRNA-seq effects. 9. Run Step21E as a separate post hoc sensitivity analysis. Identify predicted doublets within each original matrix/sample using RNA raw counts and scDblFinder, exclude them, retain frozen labels without reclustering or reannotation, and repeat aggregation and inference. This does not redefine the primary eight-candidate set. 10. Compare regenerated outputs with the supplied tables, figures, session records and MD5 manifest. Expected checkpoints are 96 contrasts, 32 comparable combinations, 18 directionally concordant combinations, eight primary candidates, 16 dataset-specific estimates, and no BH-adjusted q value below 0.05. Interpret results as exploratory and descriptive, not as a validated score, diagnostic or prognostic model, intervention-response biomarker, or causal proof.