Copy-aware deconvolution of Hi-C contacts on ecDNA. Duplicated ecDNA segments map to the same reference region, so their contacts collapse in reference space. CADET redistributes these contacts along the ecDNA path given in a BED file (Layer 1), calls enriched contacts (Layer 2) and annotates them (Layers 3–8).
| Layer | Script | Step |
|---|---|---|
| 1 | cadet_layer1.py |
copy-aware Poisson EM deconvolution |
| 2 | cadet_layer2.py |
distance-stratified background, local enrichment, FDR |
| 3 | prepare_layer3_inputs.py, cadet_layer3.py |
merge contacts into blocks |
| 4 | cadet_layer4.py |
group blocks into modules |
| 5 | cadet_layer5.py |
module priority |
| 6 | cadet_layer6.py |
genes (GENCODE v49), oncogenes (OncoKB), cCREs (ENCODE) |
| 7 | cadet_layer7.py |
module interpretation and targets |
| 8 | cadet_layer8.py, cadet_layer8_circos.py |
target recurrence vs shuffled nulls; circos plot |
requirements.txt lists the package versions we tested. With pip, install hic-straw from bioconda or build it separately. It's only needed to read .hic files.
git clone https://github.com/Biswanath2410/cadet.git
cd cadet
conda env create -f environment.yml
conda activate cadet
bash download_references.sh # OncoKB, GENCODE and ENCODE files for Layer 6ecDNA BED: one row per segment, in path order, with a header:
chrom start end orientation
chr8 127000000 127500000 +
chr8 127600000 127800000 -
Hi-C: a .hic file for the same sample.
run_full_pipeline.sh runs every layer in order. A typical run is:
BED=inputs/sample_ecDNA.bed \
HIC=inputs/sample.hic \
RES=5000 \
RUN_NAME=sample \
bash run_full_pipeline.shResults go to runs/<RUN_NAME>/ (part1/ Layer 1, part2_discovery/ and part2_fdr/ Layer 2,
cadet_final_results/ Layers 3–8).
Options: STAGES=layer6,layer7 runs only some stages, SKIP_CIRCOS=1 skips the circos plot,
Y_CACHED=runs/<run>/part1/observed_Y.txt reuses an earlier contact matrix. Each script can also be run on its own
(python scripts/<script>.py -h).
The OncoKB list changes over time; the download date is saved in reference_download_date.txt.
simulation/ has the Layer 1 benchmark (see simulation/README.md).
Chowdhury, B. (2026). Copy aware deconvolution of Hi-C interactions in extrachromosomal DNA using CADET. Research Square. https://doi.org/10.21203/rs.3.rs-11142844/v1
MIT. See LICENSE.