Cell-free DNA methylome and fragmentome analysis for relapse monitoring of Ewing Sarcoma - data processing and classifier building pipelines

This repository contains the input files and Nextflow workflows used to generate processed qseaSet objects from T7-MBD-seq data (an enrichment-based methylation capture method using a methyl-binding domain protein), and to construct the EwingSign classifier. OverviewT7-MBD-seq was applied to 86 plasma cell-free DNA (cfDNA) samples derived from Ewing sarcoma (EwS) and CIC-rearranged patients, as well as 107 non-cancer controls (NCC) (83 of which were used for classifier training). After pre-processing fastq files using the in-house T7-MBD-seq Nextflow pipeline, the resulting qseaSet objects were used within a second Nextflow pipeline for the training and application of the EwingSign classifier. This classifier is an ensemble of 25 XGBoost models trained on EwS/CIC methylation array data from Koelsche et al., 2021 (Data Reference: GSE140686). Contents of the repository: 1. Input FilesThese include all necessary input files for running both the T7-MBD-seq and EwingSign classifier Nextflow pipelines. T7-MBD-seq pipeline inputs (T7MBDseqInputFiles.zip):- starterSheet.csv — contains file paths to the FASTQ files.- sampleTable.csv — contains metadata for each cfDNA sample.- config files EwingSign classifier pipeline inputs (ClassifierInputFiles.zip):- CSV files specifying paths to array methylation data (IDAT files)- NCC qseaSet object- CpG_beta_Count_lookupTable.csv — used to convert methylation data into qseaSet objects- Contrasts.csv — defines sample groups for differentially methylated region (DMR) selection- mix.csv — defines group proportions for generating in silico mixture samples 2. Nextflow Pipelines- T7MBDseqNextflowPipeline.zip — Nextflow main workflow, modules, and R scripts for T7-MBD-seq data preprocessing- ClassifierBuildingNextflowPipeline.zip — Nextflow main workflow, modules, and R scripts for training and applying the EwingSign classifier 3. Scripts.zip- Bash scripts to run the T7-MBD-seq Nextflow pipeline for specific cfDNA sample cohorts (EwS, NCC, and NCChigherDepth)- Bash script to run the EwingSign classifier- R script to generate NCC split qseaSet objects from the original NCC qseaSet object- R script to generate CSV files containing the paths to EWS/CIC and NCC split qseaSet objects.- R and bash scripts to generate in silico dilution mixture sets (using array data from Patrizi et al., 2024; Data Reference: GSE276012) and to obtain EwingSign classifier predictions (related to Figure 3B) Code to reproduce all figures in the paper is available at https://doi.org/10.5281/zenodo.17527450 (unrestricted).