Genesis: plant DAP-seq and ATAC-seq processing

Genesis turns plant DAP-seq and paired-end bulk ATAC-seq reads into peaks, signal tracks and quality-control reports. It runs on a single server or through Nextflow on a cluster. You can also organize results in a local catalog, review metadata and QC, and export a selected dataset collection.

Start with the standalone pipeline for your assay. Study management, the catalog and AI assistance are optional.

Scientific purpose and current evidence

Genesis supplies traceable preprocessing and curated data for plant sequence-to-function prediction and experimentally testable sequence design. The chain is data → declared measurement → model input → downstream test. It is not a validated plant foundation model or an MPRA platform.

Nextflow runs the supported scientific workflows. Optional LLMs assist metadata curation and explanation; they do not grant scientific approval.

Implemented capability	Evidence scope	Evidence revision/source	Remaining acceptance
DAP and bulk ATAC workflows	Real tools on synthetic inputs; selected public subsets	Study acceptance; public comparisons, archived in b9d0cf7	Full-depth biological and task-specific acceptance
scATAC fragment ingestion, pseudobulks and model exports	Synthetic contracts and real tools/loaders on synthetic inputs	scATAC record, dated 2026-10-10 with baseline and implementation revisions	Reviewed real-data bundles, plant bias evidence and exact-bundle loader acceptance; no scATAC FASTQ processing
Read-only model-bundle inspection	Bounded artifact, convention and context-window checks	Inspection guide; report records inspector and bundle hashes	Structural checks do not approve training
Registry/review and optional workers/AI services	Offline/local-worker fixtures and mock HTTP/lifecycle tests	Validation records, each with its own tested revision	Site-specific SSH/scheduler/API/GPU acceptance and biological curation accuracy

Read the scientific overview for measurement semantics, three evidence-backed observations, and what remains between files, model training and experimental validation. Use the offline inspection example to examine a clearly labelled synthetic bundle.

Pipeline workflow maps

DAP-seq workflow

View full size · DAP-seq guide

Bulk ATAC-seq workflow

View full size · Bulk ATAC-seq guide

Install and check your environment

Install Pixi. For the default execution mode, Docker must already be available to your user. Run:

git clone https://github.com/kundajelab/genesis-preprocessing.git
cd genesis-preprocessing
pixi run install-all

This installs everything the DAP-seq pipeline needs; run it with pixi run pipeline (see DAP-seq below).

The bulk ATAC, validation and curation commands below use the genesis command. To install it, run pixi run install-all --install-cli instead, keep ~/.local/bin on your PATH (export PATH="$HOME/.local/bin:$PATH"), and check it with genesis doctor. The installer links genesis to this checkout; keep the checkout in place, or add --bin-dir /your/bin to use another command directory. See installation and upgrades for moving a checkout, using a Python wheel or removing the command.

Run your first analysis

Run the examples below from the checkout, substituting your input paths. Both pipelines need an explicitly selected reference genome; Genesis does not choose or download an assembly for you.

DAP-seq

Prepare a tab-separated sample sheet with this exact header:

sample_id	species	read1_url	read2_url	control_sample	reference_fasta
treatment	Arabidopsis_thaliana	https://example.org/treatment.fastq.gz	-	control	TAIR10.fa.gz
control	Arabidopsis_thaliana	https://example.org/control.fastq.gz	-	-	TAIR10.fa.gz

Replace the example URLs. Use - for an absent mate or control. Put the named gzipped FASTA in your reference directory. Each treatment's control must occur in the same sheet and match its species, reference and read layout.

Run the pipeline from the checkout with pixi run pipeline. Pixi runs it from the checkout folder, so give the sheet and other paths as absolute paths:

pixi run pipeline -w /scratch/genesis -n study-A /data/dap.tsv --references /data/references
# Continue the same analysis after an interruption.
pixi run pipeline -w /scratch/genesis -n study-A /data/dap.tsv --references /data/references -resume

Alternatively, call the script directly; it uses the checkout's Pixi environment itself, and relative paths resolve from your current folder:

/path/to/genesis-preprocessing/scripts/run-pipeline.sh -w /scratch/genesis -n study-A \
  dap.tsv --references /data/references

pixi run pipeline --help lists the options. Choose a profile with -p (for example -p local,conda or -p sherlock,apptainer); the default is local,docker, and a cluster is always selected explicitly. The pipeline checks the whole sheet before starting any task.

Results appear in /scratch/genesis/study-A/output/, grouped by species and sample. Treatments receive peaks and peak scores; controls receive QC and tracks. Keep the run's work/ directory to allow resume.

Peak calling uses MACS3 built from a pinned upstream revision; see MACS3 builds and resources.

Read the DAP-seq guide for complete sample-sheet rules, reference setup, analysis choices and output definitions.

Bulk ATAC-seq

Prepare one row per paired-end technical lane and a reference registry containing the FASTA, annotation-derived TSS positions, organellar contigs and checksums. The ATAC guide explains the columns and reference fields.

genesis validate inputs atac.tsv --assay bulk-ATAC --references references.json
genesis pipeline-atac atac.tsv --references references.json --outdir /scratch/atac-run
genesis pipeline-atac atac.tsv --references references.json --outdir /scratch/atac-run --resume

Results appear under /scratch/atac-run/output/. Each library has QC reports, fragments, Tn5 cut-count tracks and peaks. Technical lanes are combined within a library; biological replicates remain separate. Choose MAPQ, duplicate and adapter policies in the sample sheet before running.

Find and interpret your results

Start with the MultiQC report, then inspect the library-level QC and signal tracks. A successful run means the computation finished; it does not establish that the data are suitable for every biological analysis. In particular, FastQC flags need interpretation, and zero peaks can be a valid result.

Output	What it tells you
MultiQC and library QC	Read quality, alignment summaries and assay-specific measurements
Peaks	Regions enriched under the recorded peak-calling and control settings
bigWig tracks	Signal across the reference; check whether the track contains normalized coverage or raw cuts
DAP peak scores	Read-end RPM and mean coverage RPKM; these are different statistics
ATAC fragments and enrichment metrics	Usable paired fragments, fragment FRiP and TSS enrichment
Provenance and Nextflow reports	Inputs, analysis settings, software identity and execution records

See the DAP output definitions and ATAC outputs before comparing measurements between tools or studies.

Choose your next task

I want to…	Start here
Run DAP-seq or understand its scientific settings	DAP-seq guide
Run paired-end bulk ATAC	Bulk ATAC guide
Run on a cluster or independent servers	Execution guide
Manage a study from inputs through reviewed export	Study workflow
Validate results, search a catalog or review QC	Curation guide
Import single-cell fragments and build pseudobulks	Single-cell ATAC guide
Prepare scATAC outputs for model loaders	Model handoff
Propose metadata or explain a failure with an LLM	AI providers
Host a private model on an existing node	Model services
Compare processing methods and inspect benchmark evidence	Benchmarks
Set up a development environment or contribute	Contributor guide

The documentation index also links to reference material and validation records. Core processing, catalog queries and manual review work without AI, a GPU or an API account.

Benchmarks and tool choices

Use the benchmark guide for the experiment matrix, metric definitions, results and reproduction commands. It distinguishes package comparisons from parameter sensitivity and software tests. A higher mapping rate or more peaks alone is not evidence that one method is better.

The plant comparison report contains 64 completed Arabidopsis method/depth cases, independent TF motif evidence, biological ATAC replicate measurements, and comparisons of reference assemblies and annotation releases. Tables distinguish completed results from the ongoing A. lyrata, sorghum and maize runs. The current evidence does not establish a better replacement for the BWA-MEM2/MACS3 defaults across assays. The optimized MACS3 comparison measures caller speed, memory use and output equivalence at matched CPU allocations. The complete-workflow comparison scores Genesis and nf-core ATAC using common metric definitions.

Troubleshooting

Problem	First step
Missing input TSV from pixi run pipeline	Pixi runs from the checkout folder: give absolute paths, or call scripts/run-pipeline.sh directly.
genesis: command not found	Complete installation with --install-cli and check your PATH.
Docker or Nextflow is unavailable	Run genesis doctor; check the selected runtime and installation.
Reference or input checksum differs	Check which file changed. Preserve the old run and use a new input/run version.
A resumed run repeats some tasks	Check changed inputs, code and reference-cache paths in the Nextflow trace.
No peaks or unexpected QC values	Inspect depth, reference, assay settings and control assignment before changing parameters.
A reviewed export is empty	Read its exclusions; validation or current review decisions may be missing.
A worker is unreachable	Inspect worker status and reconnect before authorizing another attempt.