CANDiT

Code and derived data for Sinha S, Alcantara J, Perry K, et al., "CANDiT: a machine learning framework for differentiation therapy in colorectal cancer", Cell Reports Medicine 6(11):102421, 2025. https://doi.org/10.1016/j.xcrm.2025.102421

What the study does

Differentiation therapy works in some haematologic malignancies but has not translated to solid tumours, largely because tumour heterogeneity obscures the stem-cell compartment. CANDiT approaches this by building a transcriptomic network anchored on CDX2, an intestinal lineage transcription factor lost in poorly differentiated colorectal cancers, and using it to nominate nodes whose modulation reinstates lineage commitment.

The analysis prioritises PRKAB1, a regulatory subunit of AMPK, and the prediction was then tested in CRC cell lines, mouse xenografts, and a prospective cohort of patient-derived organoids. A 50-gene response signature derived from those platforms is evaluated against clinical outcome data.

Notebooks

• univariate analysis_Fig3F_5E.ipynb - univariate and multivariate regression against clinical covariates, coefficient plots, survival analysis
• multivariate analysis_Fig3E.ipynb - multivariate models on the AM cohort
• prodiff_ROC_sig.ipynb - ROC and AUC for the response signature
• corr_plot_Fig3.ipynb - correlation structure between signature components
• PRODIFF_Fig4_A_G.ipynb - panels for Figure 4
• PRODIFF_Heatmap.ipynb - expression heatmaps
• CANDiT/Prodiff-Paper.ipynb - the network construction and node ranking

Data

Derived tables used by the notebooks are committed here: cell line and xenograft differential expression (cell_xeno_deg.txt, SW480.txt, HCT114.txt), organoid results (organoid.txt, ROC_PDO.txt), in vivo results (invivo.txt), IC50 values (IC50_heatmap.txt), signature definitions (the two xlsx files) and network output (node-2(1).txt, node-3(1).txt, pgsig-res-1.txt, pgsig-res-2.txt).

Primary expression data is not included. Some notebook cells query a lab-internal expression database and will not run outside that environment; those calls are listed in MIGRATION.md.

Setup

pip install -r requirements.txt
pip install git+https://github.com/sinha7290/bioutils.git

Shared helper functions (thresholding, regression tables, survival, plotting) live in bioutils. This repository previously carried copies of lab-internal scripts that only ran against one filesystem; they have been removed. See MIGRATION.md for the mapping and for the calls that still require the internal database.

License

MIT