Disclosure: I am the author and maintainer of ExomeFlow.

I am sharing ExomeFlow, an MIT-licensed Python package for whole-exome sequencing analysis. The goal is to provide a reproducible command-line workflow from paired-end FASTQ files to annotated variant outputs, while reducing first-run setup effort.

PyPI package: https://pypi.org/project/exomeflow/

Install:

pip install exomeflow

Basic run:

exomeflow run --input-dir fastq/ --output results/

Current germline workflow

The default germline workflow includes:

• fastp quality control
• BWA-MEM alignment
• coordinate sorting and duplicate marking
• GATK BQSR
• GATK HaplotypeCaller
• hard filtering
• ANNOVAR-based functional annotation
• fastp, alignment, and GATK QC outputs, plus MultiQC summary reporting

The pipeline supports hg38 and GRCh37 reference configurations. It can retain raw and filtered variant outputs and produces per-sample annotated files by default.

Optional modes

Current opt-in functionality includes:

• Cohort joint genotyping with GenomicsDBImport and GenotypeGVCFs
• Tumor-only somatic calling with Mutect2
• Read-depth CNV calling using GATK tools
• HPO gene-to-phenotype enrichment
• InterVar-derived ACMG/AMP classification fields when the corresponding external resources are available
• Docker and Singularity/HPC use

Reproducibility and operations

ExomeFlow records completed steps for resumable runs, writes per-sample and pipeline-level logs, can process samples in parallel, and retains resolved configuration information after dependency setup.

ANNOVAR is not bundled because its licence requires individual registration. The package documentation describes this requirement and the expected reference resources.

Current limitations

• The somatic workflow is tumor-only; matched tumor-normal support is not yet implemented.
• HPO and InterVar enrichment depend on external resources and may be omitted if those resources are unavailable.
• The output is intended for research use and does not replace validated clinical diagnostic pipelines, laboratory quality systems, or expert variant interpretation.

I tested the default single-sample germline workflow on NA12878/HG001 data against GIAB NIST v4.2.1. The current documentation reports PASS-only precision of 99.41% for SNPs and 89.38% for indels, with recall of 64.67% and 66.14%, respectively. Those recall values reflect conservative hard filtering and were measured before removal of a non-standard DP < 10 site filter; I will refresh the benchmark after rerunning it with the updated filtering behavior.

I would appreciate technical feedback from people who run WES pipelines, particularly on:

1. Whether the stated default filtering and QC choices are appropriate for a general-purpose WES workflow.
2. What provenance should be mandatory in a reusable analysis report.
3. Which annotation sources, output formats, or interoperability points are most important.
4. Whether the automatic first-run dependency/reference setup is useful or creates avoidable reproducibility and deployment risks.
5. What validation or benchmarking evidence you would expect before adopting such a pipeline for research work.

Critical feedback and bug reports are welcome.