Code, summary result tables and figure sources for the manuscript:
Integrative transcriptomic and DNA methylomic profiling reveals a stroma-dominated progression program and druggable targets in phyllodes tumors
| Directory | Contents |
|---|---|
scripts/R/ |
Analysis pipeline (steps 02–07) plus _env.R path resolver and _ordinal_trend.R |
scripts/python/ |
Figure generation (08_figures.py), druggable-target selection (09_select_druggable_ruleA.py), concordance permutation test (audit_stats.py) |
scripts/bash/ |
Data download (01_download_data.sh) and pipeline orchestration (run_pipeline.sh) |
results/tables/ |
Summary result tables (differential expression, network, GO, druggability). See "Large intermediate tables" below |
results/summaries/ |
Run logs for the ordinal-trend refit and the druggable-target selection rule |
docs/manuscript/figures/ |
Manuscript figure sources (PNG + PDF, 300 dpi) |
docs/manuscript/review/ |
Methodological provenance: the audit report and the numbers ledger that maps every reported value to its source file |
What is intentionally not published here: the manuscript text and the submission package. This repository publishes the analysis, not the article; the peer-review process is not pre-empted. The numbers ledger documents the provenance of every value reported in the manuscript without reproducing the manuscript itself.
All four datasets are public and were retrieved from NCBI GEO via GEOquery:
| Dataset | Platform | n | Contrast(s) used |
|---|---|---|---|
| GSE231574 | Illumina HumanMethylationEPIC | 33 | Malignant 18 / Borderline 9 / Benign 2 / FA 2 / Metaplastic 2 |
| GSE179458 | Illumina HumanMethylation450K | 96 | PT subset 33 (Malignant 10 / Benign 15 / Borderline 8) |
| GSE235267 | Agilent lncRNA + mRNA | 16 | FA 4 / Benign 6 / Malignant 6 |
| GSE78071 | Illumina HumanHT-12 | 48 | FA 24 / Benign 16 / Borderline 4 / Malignant 4 |
No new data were generated. All samples are de-identified public data.
# R 4.3.3 / Bioconductor 3.18
# minfi, DMRcate, limma, GEOquery, ConsensusClusterPlus,
# org.Hs.eg.db, IlluminaHumanMethylation{EPIC,450k} annotation packages
# Python >= 3.10
# pandas numpy scipy matplotlib networkxCreate a .env file in the repository root:
DATASET_PATH='E:\path\to\Phyllodes tumor'
DATASET_PATH_TEMP='F:\path\to\temp'All scripts resolve paths through scripts/_env.py (Python) or scripts/R/_env.R (R).
Both abort with a clear error if DATASET_PATH does not exist — they never fall
back silently to the source tree.
Layout expected under DATASET_PATH: raw/, data/processed/, results/tables/, results/figures/.
bash scripts/bash/01_download_data.sh # checkpointed GEO download
bash scripts/bash/run_pipeline.sh # steps 02 -> 06
Rscript scripts/R/_ordinal_trend.R # continuous ordinal severity models
python scripts/python/09_select_druggable_ruleA.py # druggable target selection
python scripts/python/08_figures.py # all manuscript figures
python scripts/python/audit_stats.py # concordance permutation testExpression normalisation (GSE78071). The GEO series matrix contains no Illumina
negative-control probes (GPL14951 Probe_Type is entirely A/I/M/S, 0 control rows), so
limma::neqc()'s normexp background correction is not applicable. Intensities were
therefore log2-transformed and quantile-normalised between arrays
(limma::normalizeBetweenArrays(method = "quantile")).
Ordinal severity model. The trend analysis fits histological grade as a continuous
covariate (benign = 1, borderline = 2, malignant = 3) inside the same four-group linear
model used for the pairwise contrasts, so residual degrees of freedom and variance
structure are identical. An earlier contrast written as
(borderline − benign) + (malignant − borderline) was discarded: it is algebraically
identical to malignant − benign and would have duplicated that result.
Druggable target selection. Targets are selected by an explicit, fully reproducible rule applied to all 851 candidates:
adj.P.Val(PT vs FA) < 0.05 AND PT-mechanism annotation AND >= 1 curated candidate agent
This yields eight targets with no manual nomination step. Note that the differential-expression support is entity-level (PT vs fibroadenoma): none of the eight reaches significance in the malignant-vs-benign grade contrast. Approved direct inhibitors exist for KIT and PDGFA only.
Probe-level DMP tables (GSE231574, 519,767 probes × 6 contrasts; GSE179458, 361,964
probes) are not stored in this repository: together they exceed 460 MB. They are
intermediate products that are fully regenerated by scripts/bash/run_pipeline.sh and
scripts/R/_ordinal_trend.R from the public GEO inputs listed above.
The summary tables required to verify every number reported in the manuscript are
included under results/tables/.
docs/numbers_ledger_*.md is the authoritative record of every quantitative value
reported in the manuscript, together with the source file and the reporting constraints
for each section. If there is ever a discrepancy between the manuscript text and this
repository, the ledger and the CSV files here take precedence.
Code: MIT (see LICENSE).
Result tables and figures: CC BY 4.0.
See CITATION.cff. Please also cite the original data-generating publications for the
four GEO datasets.