Pathway-Level Convergence Despite Gene-Level Heterogeneity in Skeletal Muscle from Patients with Cancer Cachexia: Cross-Cohort Analysis Code and Derived Results

This research compendium contains analysis code, derived expression objects, complete statistical result tables, software session information, interpretation summaries, and publication-quality figure files supporting a cross-cohort robustness analysis of human skeletal-muscle transcriptomes in cancer cachexia. The study reanalysed two public rectus-abdominis cohorts from the NCBI Gene Expression Omnibus, GSE130563 and GSE85017. The workflow included independent RAW-CEL processing of GSE85017, cohort-specific limma modelling, CAMERA and fgsea Hallmark testing, complete leave-one-out analyses, cross-cohort leading-edge comparison, meta-analysis of 21,325 common genes, and sensitivity analysis of nine cellular marker modules. The principal finding is that restricted pathway-level convergence can persist across independent clinical cohorts despite weak overall concordance, no shared FDR-significant gene, no statistically significant shared leading-edge core, method dependence, sensitivity to individual patients, and no reproducible cellular-marker-module shift. These results support an important distinction between replication of a pathway label and replication of the underlying molecular or mechanistic architecture. Six Hallmark pathway labels met the predefined rank-based replication criteria, with UV response down and androgen response showing the greatest patient-level stability. Raw source files are not redistributed in this archive and remain publicly available from NCBI GEO under accession numbers GSE130563 and GSE85017. The original GSE85017 RAW-CEL processing script and its Windows launchers are included in this repository, together with derived outputs, execution logs, software session information, complete result tables, and figure files. Analysis code is licensed under the MIT License. Derived tables, figures, and documentation are licensed under the Creative Commons Attribution 4.0 International licence.