Reproducibility dataset for "Sampling context and tissue composition contribute to stromal transcriptomic changes after neoadjuvant chemotherapy in breast cancer"

Abstract

This repository contains the complete reproducible computational workflow supporting the study “Sampling context and tissue composition contribute to stromal transcriptomic changes after neoadjuvant chemotherapy in breast cancer.” The study is a retrospective patient-level analysis of four publicly available longitudinal breast cancer transcriptomic datasets from the NCBI Gene Expression Omnibus (GEO): GSE191127, GSE21974, GSE32072, and GSE32603. The final analytical dataset comprises 105 treated patient pairs and 4 untreated paired controls. The repository provides the scripts, configuration files, processed analytical matrices, predefined transcriptional signatures, patient-level scores, statistical outputs, figure source data, computational figures, robustness analyses, and reproducibility documentation required to independently reconstruct the reported analyses. The workflow is organized into five sequential stages: Sample identification and PRE/POST pairing, including explicit inclusion and exclusion rules. Cohort-specific expression reconstruction and gene-level standardization from the original GEO datasets. Calculation of transcriptional signature scores and derived variables, including PROLIF, ECM, FIBROBLAST_LINEAGE, STEM_MES, and the Treatment Remodeling Index (TRI), together with sensitivity and robustness analyses. Statistical analyses and computational figure generation, including longitudinal comparisons, correlation and regression analyses, exact permutation testing, and exploratory recurrence-free survival analyses. Independent reproducibility audit, linking the frozen computational outputs to the numerical results reported in the manuscript and testing end-to-end reconstruction of the analysis. The final reproducibility audit evaluates 170 numerical claims from the manuscript, with 170 classified as EXACT at the reported precision, 0 as ROUNDING, and 0 as DIFFERENT. An independent clean-run test starting from the original public GEO inputs regenerated 44 of 44 evaluated outputs from Steps 1–4 byte-identically. Primary transcriptomic data are not redistributed in this archive and remain available from their original GEO records. The repository instead contains the code and configuration required to retrieve and reconstruct these data, together with the processed data and analytical outputs underlying the study. The analyses were developed and validated using Python 3.9.13, with software dependencies specified in the accompanying requirements.txt file. Detailed methodological decisions, frozen analytical definitions, file inventories, SHA-256 checksums, source data for Figures 1–4, and instructions for independent reproduction are included in the repository. The results support the interpretation that proliferative suppression after neoadjuvant chemotherapy is comparatively reproducible across treated cohorts, whereas ECM- and fibroblast-associated transcriptomic changes are more dependent on cohort, tissue composition, and sampling context. Accordingly, TRI is treated as a cohort-dependent descriptive index rather than a validated treatment-response biomarker. This study was supported by Proyecto UTA MAYOR 2026 N°4779-26.