2026-10-08T08:06:18Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33849385 2026-09-16T13:19:59Z category_24784 category_24208 category_24310 category_29872 item_type_3 month_year_09_2026

Analysis release: early-onset and late-onset colorectal cancer differ by a graded transcriptional shift rather than a discrete regulatory state Jorge Mata-Garrido (24666928) Cancer genetics Bioinformatics and computational biology not elsewhere classified Genomics Biostatistics Early-onset colorectal cancer (EO-CRC) transcriptional states attractor models Cell-type composition study design challenges dichotomisation reanalysis/reinterpretation reproducibility benefit <p dir="ltr">Analysis code, derived data and results accompanying the study "Early-onset and late-onset</p><p dir="ltr">colorectal cancer differ by a graded transcriptional shift rather than a discrete regulatory</p><p dir="ltr">state" (Desterke, Yu & Mata-Garrido).</p><p><br></p><p dir="ltr">The study reassesses whether a 15-gene CBX3-associated programme separates early-onset (EOCRC)</p><p dir="ltr">from late-onset (LOCRC) colorectal tumours into discrete transcriptional states, as opposed to</p><p dir="ltr">shifting their location within a single distribution. The two readings are usually supported by</p><p dir="ltr">the same evidence - differential expression, clustering displays, low-dimensional embeddings -</p><p dir="ltr">but they are different claims, and only the second is established by those analyses.</p><p><br></p><p dir="ltr">Across 98 tumours from a public discovery cohort, the programme shifts by roughly half a standard</p><p dir="ltr">deviation at module level, yet carries almost no geometric structure: the silhouette coefficient</p><p dir="ltr">of the onset labels is 0.034 against a permutation null of 0.000, leave-one-out discrimination</p><p dir="ltr">reaches an area under the curve of 0.628, and Gaussian mixture models prefer a single component</p><p dir="ltr">for four of five module scores. Between a quarter and seventy per cent of the module effects</p><p dir="ltr">disappear after adjustment for leukocyte, stromal and epithelial composition proxies. The cohort</p><p dir="ltr">contains no patient aged between 50 and 69 years, so every threshold between those ages yields an</p><p dir="ltr">identical partition and identical statistics, and a discrete boundary cannot be distinguished from</p><p dir="ltr">a continuous age gradient by this design.</p><p><br></p><p dir="ltr">Contents:</p><p dir="ltr">- analysis.py, figures.py, figures_supp.py - regenerate every number, table and figure in one pass</p><p dir="ltr">- rdata.py - minimal reader for R's RDX3 serialization format, so no R installation is required</p><p dir="ltr">- ols.py - ordinary least squares with HC3 robust covariance, and Benjamini-Hochberg correction</p><p dir="ltr">- data/ - inputs, with provenance and the original R analysis retained for reference</p><p dir="ltr">- results/ - all result tables, per-sample module scores, intermediate arrays and summary.json</p><p dir="ltr">- figures/ - main and supplementary figures at 300 dpi</p><p dir="ltr">- CHECKSUMS.txt, environment.txt - SHA-256 of every file, and interpreter and package versions</p><p><br></p><p dir="ltr">A single seed (20260916) governs all stochastic components. Two negative controls are included</p><p dir="ltr">deliberately and reported in the manuscript because both came out negative: the group separation</p><p dir="ltr">on a t-SNE projection is reproducible across 200 seeds, and standardising features across the full</p><p dir="ltr">cohort before cross-validation changes the area under the curve by 0.0004.</p><p><br></p><p dir="ltr">Underlying data: NCBI Gene Expression Omnibus accession GSE213092. No new data were generated.</p> 2026-09-16T13:19:59Z Dataset Dataset 10.6084/m9.figshare.33849385.v1 https://figshare.com/articles/dataset/Analysis_release_early-onset_and_late-onset_colorectal_cancer_differ_by_a_graded_transcriptional_shift_rather_than_a_discrete_regulatory_state/33849385 CC BY 4.0