Use Harako-RNAseq when
- You need broad access from Windows or Linux.
- Your main goal is Salmon, tximport, and DESeq2.
- You do not need BAM output.
Public research alpha · v0.1.0-alpha.1
Harako-GPU
GPU-assisted BAM generation and reproducible, version-pinned RNA-seq quantification in one GUI
Harako-GPU is a local RNA-seq workflow validated on Ubuntu 24.04 with an RTX 3090 and 128 GB RAM. Each run records and pins the workflow, containers, reference data, indices, and analysis parameters.
FASTQ → fastp → Parabricks STAR → BAM / junction / GeneCounts + Salmon → matrices / MultiQC
source-availableUbuntu + NVIDIA GPUresearch use onlynot for diagnostic use
GPU-assisted edition
A compact Harako-native workflow
Harako-native uses a compact Nextflow workflow for fastp preprocessing, Parabricks alignment, FeatureCounts, and MultiQC. The Harako controller then runs Salmon quantification, expression-matrix generation, concordance analysis, output validation, and support-information collection.
- fastp preprocessing
Prepares the paired-end FASTQ files used by downstream steps. - Parabricks alignment
Runs the selected one-pass or two-pass alignment profile. - Alignment results
Produces BAM, splice-junction, and GeneCounts files. - FeatureCounts
Performs biotype-level QC withexonandgene_type. - Salmon and MultiQC
Produces version-pinned quantification and an integrated QC report.
New plans on Ubuntu systems with a validated environment profile use Harako-native by default. The full nf-core/rnaseq 3.26 workflow remains available as an expert reference backend. Windows/WSL plans and historical plans continue to use the reference backend by default.
Who is Harako-GPU for?
For laboratories that already operate a compatible NVIDIA GPU workstation, Harako-GPU brings BAM generation, splice-junction analysis, GeneCounts, Salmon quantification, matrix generation, and reporting into one reproducible run.
Which Harako should I use?
| Question | Harako-RNAseq (CPU) | Harako-GPU alpha |
|---|---|---|
| Environment | Supports a broader range of Windows and Linux systems | Tested on Ubuntu 24.04 with an RTX 3090 and 128 GB RAM |
| Primary workflow | Salmon → tximport → DESeq2 or QC-only analysis | GPU-assisted BAM, junction, and GeneCounts generation with Salmon |
| BAM / junction / GeneCounts | Not part of its primary workflow | Available through tested one-pass and two-pass profiles |
| Salmon | Designed for general quantification | 2.5.1 is the default; 1.10.3 is available for compatibility |
| DESeq2 | Available in the CPU edition | Not implemented |
| Hardware | Designed primarily for CPU-based Docker execution | Tested on the RTX 3090 / 128 GB host class |
| Maturity | Public beta | Public research alpha |
| Recommended user | Researchers who need quantification and DESeq2 across a broader range of systems | Researchers who already have a compatible NVIDIA GPU and need BAM, junction, or GeneCounts outputs |
Use Harako-GPU alpha when
- You already operate a validated NVIDIA GPU workstation.
- You need BAM, splice-junction, and GeneCounts outputs.
- You need to select one-pass or two-pass alignment, or compare pinned Salmon versions.
Harako-GPU does not replace Harako-RNAseq. The two products run different workflows, so no matched CPU speedup is claimed.
What has been validated
Complete analysis of the public human dataset
One-pass and two-pass analyses completed successfully for all 36,349,964 paired fragments in the public ERR188044 dataset. BAM integrity, expression matrices, MultiQC reports, and archived results were also checked.
Agreement with the reference backend
For the fixed C1 input, Harako-native one-pass and two-pass runs matched the reference backend for the principal analysis results and produced identical BAM record streams.
GUI validation
The Ubuntu GUI completed C1 one-pass and two-pass runs from plan creation through result display, output validation, and support-information collection. Application-level and HTTP tests also passed. A compatible browser was unavailable on the Ubuntu test system, so final interaction testing in an actual Ubuntu browser was not performed.
See the validated observations and their limitations →
The published validation records show that the workflow behaved as expected for the documented environment, settings, and input data. They do not establish biological validity, cross-host equality, support for arbitrary hardware, or a matched CPU speedup.