Edi-Set-Flow - Detecting RNA editing at scale

A robust pipeline for RNA editing detection and differential analysis in bulk RNA-seq

This is a beta release - please try it and report any issues here

Overview

Edi-Set-Flow runs in five stages:

1. Setup — Downloads reference genome, GTF, REDIportal (hg38 only), and common variant exclusion lists (dbSNP/gnomAD), then builds aligner indices. All resources are cached in esf_resources/ and reused across runs.
2. Alignment — Optionally trims reads with fastp, aligns to the reference (STAR or BWA-MEM2), filters BAMs, estimates per-sample coverage with mosdepth, and automatically detects library strandedness.
3. Site Discovery & Counting — JACUSA2 scans for candidate A-to-I (A>G) RNA editing sites in two passes: a loose discovery pass to identify candidate sites, followed by a stricter counting pass to quantify allele counts. Both passes run in parallel across samples.
4. Annotation — Ensembl VEP annotates variant consequences (gene, transcript, impact); vcfanno adds REDIportal catalog membership and repeat-masker context.
5. Reporting & Statistics — MultiQC aggregates alignment and QC metrics; the companion edisetr R package fits per-site GLMs across experimental groups and generates an interactive HTML report.

Getting Started

1) Requirements

• Nextflow (≥ 22.03.0)
• Container engine, any of:
  • Apptainer
  • Singularity
  • Docker

2) Sample Manifest (CSV format)

• Required Columns:
  • sample_id — unique identifier for each sample
  • group — experimental condition used as the primary contrast in the GLM (e.g. tissue, treatment)
  • One of the following input options per sample:
    • fastq1 (+ optional fastq2) — paths to local FASTQ files; omit fastq2 or leave blank for single-end data
    • run_accession — SRA or ENA run accession (e.g. SRR..., ERR...); reads are downloaded automatically using fasterq-dump
• Optional Columns:
  • Arbitrary covariates to include in the GLM as fixed effects (specified via --report_fixed_effects)
  • Columns containing numbers are treated as numeric; all others are treated as factors
• Example (local FASTQs):

  sample_id,fastq1,fastq2,group,sex,age
  SRR1311086,/PATH/TO/SRR1311086_1.fastq.gz,/PATH/TO/SRR1311086_2.fastq.gz,cortex,male,50
  SRR1477080,/PATH/TO/SRR1477080_1.fastq.gz,/PATH/TO/SRR1477080_2.fastq.gz,cerebellum,female,60
  SRR1085825,/PATH/TO/SRR1085825_1.fastq.gz,/PATH/TO/SRR1085825_2.fastq.gz,hippocampus,male,50
  ...
  

• Example (SRA/ENA accessions):

  sample_id,run_accession,group,age
  SRR5961804,SRR5961804,CTRL,38
  SRR5961807,SRR5961807,CTRL,55
  SRR5961808,SRR5961808,MDD,38
  ...
  

3) Run Edi-Set-Flow

• Example:

  nextflow run bahlolab/Edi-Set-Flow \
      -revision 26.03-beta.2 \
      -profile hg38,singularity \
      -resume \
      --input sample_manifest.csv \
      --report_fixed_effects sex,age \
      --outdir esf_results

• Quick test run:

  The test profile provides a built-in manifest of 10 SRA samples (5 CTRL / 5 MDD, human prefrontal cortex). Reads are downloaded automatically via fasterq-dump, so no input files are needed:

  nextflow run bahlolab/Edi-Set-Flow \
      -revision 26.03-beta.2 \
      -profile hg38,singularity,test \
      -resume \
      --outdir esf_test_results

• Notes:

  • Profiles:
    • Genome: hg38, mm10 or mm39 are supported. Other genome builds require custom specification — see nextflow.config
    • Container Engine: singularity, apptainer, or docker
    • Examples: -profile hg38,singularity or -profile mm10,docker
  • Resources (e.g. reference genome) are downloaded and stored in esf_resources/; use --resource_dir /path/to/resource_dir to share a single cache across multiple runs or users
  • REDIportal annotation is only supported for hg38

Parameters

Key parameters grouped by function. Defaults are set by the -profile where noted.

Input / Output

Parameter	Default	Description
--input	required	Path to sample manifest CSV
--outdir	output	Directory for pipeline outputs
--resource_dir	esf_resources	Reference data cache directory; share across runs to avoid re-downloading

Preprocessing

Parameter	Default	Description
--aligner	BWAMEM2	Aligner to use: BWAMEM2 or STAR. STAR is recommended when splice-aware alignment is important
--fastp_mode	off	Read QC/trimming: omit to skip, MINIMAL (adapter trimming only), STRICT (quality filtering + adapter trimming)
--drop_duplicates	true	Remove PCR duplicates from BAMs before site detection

Site Detection

Parameter	Default	Description
--min_mapq	20	Minimum mapping quality passed to JACUSA2
--infer_enzyme	true	Infer editing enzyme type from data (ADAR: A>G; APOBEC: C>T)
--n_intervals	10	Number of genomic intervals for parallelisation — increase for larger datasets or more available CPUs

GLM / Report

Parameter	Default	Description
--report_fixed_effects	—	Comma-separated covariate columns from the manifest to include in the GLM (e.g. sex,age). group is always included automatically
--report_model	quasibinomial	GLM family: quasibinomial (recommended — handles overdispersion), binomial, linear, or arcsine
--report_min_med_dp	10	Minimum median depth across samples for a site to appear in the report
--report_adar_only	true	Restrict the report to A>G (ADAR-type) sites only

Output Files

File	Description
EdiSetFlow.report.html	Interactive Edi-Set-Flow HTML report — see example (GTEx brain)
EdiSetFlow.sample_counts.csv.gz	Reference and alternate allele counts per site and sample
EdiSetFlow.sample_summary.csv.gz	Summary statistics (median depth & editing rate) per sample
EdiSetFlow.site_summary.csv.gz	Per-site statistics and annotation (e.g. gene, region, REDIportal membership, VEP consequence)
EdiSetFlow.glm_summary.csv.gz	GLM coefficients, standard errors, and significance per site
EdiSetFlow.glm_anova.csv.gz	ANOVA test results for each model term per site
EdiSetFlow.glm_contrasts.csv.gz	Pairwise group comparisons for differential editing
EdiSetFlow.glm_margins.csv.gz	Estimated marginal editing rates per group
multiqc_report.html	MultiQC summary report (fastp, STAR/BWA-MEM2, mosdepth, etc.)

Running on HPC & Cloud

Nextflow handles job submission and resource management, but a few settings are worth configuring for cluster or cloud environments.

Executor (SLURM, PBS, AWS, etc.)

By default Nextflow runs processes locally. To submit jobs to a cluster scheduler or cloud provider, add an executor block to a local nextflow.config in your working directory, or pass it with -c custom.config:

process {
    executor = 'slurm'       // or 'pbspro', 'awsbatch', 'google-lifesciences', etc.
    queue    = 'your-queue'
}

See the Nextflow executor documentation for all supported platforms and options.

Overriding CPU / Memory per Process

Each tool runs under a named label (e.g. jacusa2, star, vep). Override resources for any label in your config:

process {
    withLabel: 'jacusa2' { cpus = 16; memory = '80 GB' }
    withLabel: 'star'    { cpus = 8;  memory = '40 GB' }
}

Resuming Failed Runs

Always pass -resume when restarting a run. Nextflow caches completed tasks and will skip them, restarting only from the point of failure.

Container Cache

Set the appropriate environment variable so containers are pulled once and shared across runs:

export NXF_APPTAINER_CACHEDIR=/shared/path/apptainer_cache   # Apptainer / Singularity
export NXF_SINGULARITY_CACHEDIR=/shared/path/singularity_cache

Sharing Reference Resources

Pass --resource_dir /shared/path/esf_resources to point all runs at a shared reference cache, avoiding repeated downloads of the genome, GTF, and annotation databases.

Debugging Failed Runs

• Pipeline-level errors (e.g. invalid parameters, missing files, config problems) are logged to .nextflow.log in your working directory. Check the bottom of this file first.

• Task-level errors (e.g. a tool crashing mid-run) are captured in the task's work/ subdirectory. Nextflow prints the work directory path in the error output — navigate there and inspect:

  • .command.sh — the exact shell command that was run
  • .command.log — stdout/stderr from the task
  • .exitcode — the exit code

  Common fixes are increasing the CPU/memory or walltime allocation for the failing process (see Overriding CPU / Memory above).

Getting Help

• Nextflow documentation: nextflow.io/docs
• Pipeline issues and questions: GitHub Issues

References

Piechotta, M., Naarmann-de Vries, I. S., Wang, Q., Altmüller, J. & Dieterich, C. (2022) RNA modification mapping with JACUSA2. Genome Biology, 23(1), 115.

D'Addabbo, P., Cohen-Fultheim, R., Twersky, I., Fonzino, A., Silvestris, D. A., Prakash, A., … & Picardi, E. (2025) REDIportal: toward an integrated view of the A-to-I editing. Nucleic Acids Research, 53(D1), D233–D242.

McLaren, W., Gil, L., Hunt, S. E., Riat, H. S., Ritchie, G. R., Thormann, A., … & Cunningham, F. (2016) The Ensembl Variant Effect Predictor. Genome Biology, 17, 1–14.

da Veiga Leprevost, F., Grüning, B. A., Alves Aflitos, S., Röst, H. L., Uszkoreit, J., Barsnes, H., … & Perez-Riverol, Y. (2017) BioContainers: an open-source and community-driven framework for software standardization. Bioinformatics, 33(16), 2580–2582.

Di Tommaso, P., Chatzou, M., Floden, E. W., Barja, P. P., Palumbo, E. & Notredame, C. (2017) Nextflow enables reproducible computational workflows. Nature Biotechnology, 35(4), 316–319.