Skip to content

DecoNFlow

Computational deconvolution of DNA methylation data, with reference-based and reference-free tools in one Nextflow pipeline.

Get started Usage

Nextflow run with docker run with singularity nf-test

DecoNFlow estimates the cell type composition of bulk DNA methylation samples. It selects differentially methylated regions (DMRs) from reference samples, runs reference-based and reference-free deconvolution tools, and compares their estimates in an HTML report, including an evaluation against the expected proportions when they are known. It also allows benchmarking all the tools included in the pipeline.

  • Quick start

    Prepare the samplesheets, pick the tools and run the pipeline with Docker or Singularity.

    Quick start

  • Usage

    Reference and bulk samples, coverage files, reference matrices, regions and ground truth.

    Usage

  • Parameters

    Every parameter of the pipeline, generated from its parameter schema.

    Parameters

  • Output

    The combined proportions of all tools, the HTML report and the intermediate files.

    Output

Pipeline overview

DecoNFlow metro map

Click the map to open the interactive version, which you can pan, zoom and filter by line.

Step Tools
DMR selection limma
Reference-based deconvolution, beta values meth_atlas, CIBERSORT, EpiDISH, Houseman's CP (equality), Houseman's CP (inequality), EpiSCORE, MethylResolver, PRMeth
Reference-based deconvolution, read counts CelFiE, MetDecode
Reference-free deconvolution MeDeCom, RefFreeCellMix

The methylation data of the reference and bulk samples can be provided as Bismark coverage files (.cov.gz) or MethylDackel bedGraph files (.bedGraph.gz).

Quick start

New to Nextflow?

See this page on how to set up Nextflow (>= 26.04.0) and Docker or Singularity.

  1. Prepare a samplesheet with the bulk samples to deconvolve and, for reference-based deconvolution, a samplesheet with the reference samples.
  2. Select at least one deconvolution tool (e.g. --meth_atlas). Reference-based tools need the reference samples (--input) together with a regions file (--regions) for the limma DMR selection, or a reference matrix (--ref_matrix). CelFiE and MetDecode need read counts instead of beta values, so they use the reference samples or their own reference matrix (--celfie_ref_matrix, --metdecode_ref_matrix).

  3. Write the parameters in params.yaml:

    params.yaml
    input: ./resources/reference.csv
bulk_samples: ./resources/bulk_samples.csv
regions: ./resources/regions.bed
outdir: ./results
meth_atlas: true

  4. Run the pipeline:

    === "Docker"

      ```bash
  nextflow run VIBTOBIlab/DecoNFlow -params-file params.yaml -profile docker
  ```

    === "Singularity"

      ```bash
  nextflow run VIBTOBIlab/DecoNFlow -params-file params.yaml -profile singularity
  ```

Allocate at least 8 CPUs and 16 GB of RAM to run the pipeline. The results are described on the output page.

Credits

The scripts and containers have been written and built by Edoardo Giuili (@edogiuili), Maísa Santos (@mferrodo) and Sofie Van de Velde (@sofvdvel), who are also the maintainers.

Citation

If you use this pipeline for your analysis, please cite:

A benchmark of DNA methylation deconvolution methods for tumoral fraction estimation using DecoNFlow

Edoardo Giuili, Sofie Van de Velde, Sam Kint, Maísa R Ferro dos Santos, Lotte Cornelli, Sofie Roelandt, Kathleen Schoofs, Renske Imschoot, Ruben Van Paemel, Leander Meuris, Celine Everaert, Katleen De Preter.

bioRxiv 2025 Nov 27. doi: 10.1101/2025.11.27.688590

The tools used by the pipeline are listed on the citations page.