R-loop-related prognostic signature in colorectal cancer: data-analysis pipeline

Abstract

Data-analysis pipeline for a cross-cohort study of an R-loop-related gene-expressionprogramme in colorectal cancer, built entirely on public data. The pipeline constructs a CRC-specific R-loop candidate gene set, fits a penalised Cox model inTCGA-COAD/READ and evaluates the resulting score in five independent GEO cohorts (per cohort,pooled, and stage-adjusted), characterises what the score tracks (mismatch-repair status,tumour purity, proliferation, immune and cytolytic-activity programmes, single-celldetectability), maps the programme onto single-cell and spatial-transcriptomic compartments,derives consensus molecular subtypes of the candidate gene set with attribution and CMS/CRIScomparisons, predicts drug response from GDSC2 with oncoPredict, and stress-tests every layerwith sensitivity, stability, calibration and permutation-null analyses, including an end-to-endlabel permutation through the whole pipeline and a pooled-external null model fitted on randomgene sets of the same size. The archive contains the data-analysis code (67 R and Python scripts) together with a README,a SHA256 manifest and a snapshot of the software environment. Each public input is listed inthe README with its accession, and the pipeline retrieves and normalises every dataset itself,so the analysis reproduces from the original sources without data redistribution.