A compact Nextflow workflow with version-pinned analysis settings

Harako-native workflow

1. fastp preprocessing
   Uses the specified container to prepare the paired-end FASTQ files used by downstream steps.
2. Parabricks alignment
   Uses the fixed one-pass or two-pass settings and the computational resources assigned to that profile. The technical process ID is PARABRICKS_RNA_FQ2BAM.
3. Recording alignment outputs
   Records each output's role, sample, relative path, file size, and SHA-256 checksum in a backend-independent manifest.
4. FeatureCounts biotype-level QC
   Uses -t exon -g gene_type for GENCODE 49.
5. MultiQC report
   Summarizes the defined QC results while retaining access to the underlying metrics.

Salmon is not run inside Nextflow. The Harako controller uses the processed paired-end FASTQ files, an explicitly selected library type, and six threads. Quantification therefore runs only once for each selected Salmon profile.

Alignment profiles

To preserve reproducibility, users cannot change the analysis-tool parameters or allocated resources freely. Harako-GPU does not increase memory dynamically or switch automatically to CPU STAR.

Counts and Salmon quantification

For the fixed GRCh38.p14 / GENCODE 49 reference pack, FeatureCounts uses feature type exon and group type gene_type. This provides biotype-level QC and remains separate from STAR ReadsPerGene.out.tab GeneCounts.

Salmon 2.5.1 is the default for new analyses. Salmon 1.10.3 remains available for compatibility with earlier analyses. Repeated Salmon 1.10.3 runs were highly concordant, but the numerical results were not identical.

The compare-both mode runs both versions sequentially on the same processed FASTQ files and compares the abundance differences and rank concordance.

Differences caused by the software version, implementation, index, and execution conditions can be confounded with biological differences between experimental groups. This concern is not limited to floating-point rounding.

Harako-GPU therefore records and pins the Salmon version, container image, index, options, and library type for every run. Results produced with different Salmon versions should not be combined without explicitly accounting for the version difference.

Analysis plans, execution, and resume

Before a run starts, its analysis plan records the workflow, profiles, containers, reference data, output settings, computational resources, execution environment, and approval hash. Preparation checks these conditions again. During a run, persistent process, lock, and state records are used to manage execution and resume. A completed run cannot be resumed.

The GUI uses the same planning and execution services as the command-line interface for detailed output validation and support-information collection.

Reference backend for comparison

The nfcore_rnaseq_3_26_reference backend preserves a specific nf-core/rnaseq 3.26 configuration for comparison and expert use. The standard Harako-native workflow does not use the additional plugin and large container-image set required by this reference backend.

Limitations

• This is a public research alpha for research use only. It is not intended for diagnostic or clinical use.
• Testing covers only the documented environment and reference data. Other RTX systems, memory capacities, custom references, and arbitrary FASTQ datasets are not guaranteed to work.
• CPU STAR and DESeq2 are unavailable. No matched CPU STAR benchmark is provided, and not every workflow stage uses the GPU.
• Automatic archiving and deletion are not implemented.
• A compatible browser was unavailable on the Ubuntu test system, so final interaction testing in an actual Ubuntu browser was not performed. Application-level tests, HTTP checks, and C1 one-pass and two-pass analyses were completed.
• Exact equality across different hardware systems and agreement with biological ground truth were not evaluated.