Roadmap · interactive

The SeqDesk pipeline roadmap

SeqDesk2026-06-30Interactive

SeqDesk runs analysis pipelines on a sequencing facility’s own server. This page is an honest look at the shelf: what runs today, what we add next and why, and the low-effort work that makes existing results more useful. It is framed as where we are and what’s next — not a finished platform. Scroll, and each part builds one step at a time.

Today the shelf holds 7 pipelines at three different stages. The honest way to read a roadmap is to see the whole shelf first — including the parts that aren’t finished — and only then talk about what moves next.

The shelf today7 pipelines · from the store

1available

3beta

3planned

MAG Pipelinemag

available

Taxonomic Profiling (Kraken2 + Bracken)kraken2-bracken

beta

Study MultiQCmultiqc

beta

NanoPlot (Long-read QC)nanoplot

beta

Taxonomic Profilertaxprofiler

planned

Amplicon Analysisampliseq

planned

Bacterial Assemblybacass

planned

Seven workflows, one shelf — not all at the same stage, and the post won’t pretend they are.

Seven workflows, one shelf

SeqDesk ships seven analysis pipelines today. They are not all at the same stage, and this post won't pretend they are — one is production, three are beta, three are planned. Start by seeing them all in one place.

What you can actually run

Four of the seven run on real data right now: the MAG pipeline plus three in beta — Kraken2 + Bracken, Study MultiQC, and NanoPlot. Beta means it works and is wired in, not that it is finished.

One is production-grade

Only the mag pipeline is marked available. It assembles, bins, and evaluates metagenomes in one workflow — contigs, genome bins, bin-quality metrics, taxonomy, a MultiQC report. It is the anchor everything else is measured against.

Grouped by the job, not the tool

Sorted by purpose: three metagenomics workflows, two QC workflows, one amplicon, one bacterial-genome. The shape of the shelf already shows where SeqDesk is strong and where it is thin.

Honest labels

The three planned pipelines — taxprofiler, ampliseq, bacass — have names and outputs defined, but no runnable code in the app yet. We keep them dim. The rest of the post is about turning that column gold.

So why does the planned column move at all? Because in SeqDesk, adding a pipeline is the cheap part. The expensive part is making what comes out of it useful.

Why adding one is cheapdeclared, not coded per run

samplesheetprocessoutputs

5from the sample

2from the facility

0you type

your server · behind your firewall

databasestorage+ new pipeline

data never leaves →|

cost isn’t wiringit’s making outputs usable

A pipeline is a declaration

SeqDesk runs Nextflow workflows. Adding one means describing its inputs, parameters, and outputs — not writing bespoke glue for every run. That is why the planned column can move at all.

The sample already knows its fields

SeqDesk fills a run’s samplesheet from records the facility already entered. For a MAG launch, all five samplesheet columns come from the sample’s own records and two reference-DB params are set once by the facility — so the researcher types zero. Every new pipeline inherits that autofill.

The data never leaves the building

SeqDesk runs on the facility's own server. A run reads inputs and writes outputs on your machines; the database is not shipped anywhere. Adding a pipeline does not add a place your data has to travel to.

So the cost isn't wiring

If declaring a workflow is cheap and the data stays put, the real bottleneck is what comes out: are the results browsable, comparable, submittable? That is where the roadmap should spend effort — and it reframes the rest of this post.

With that in mind, here are the three planned pipelines — in the order we intend to build them, each with the one number that argues for it.

What’s next, in order3 planned · each with its reason

1
Bacterial assemblybacassReads pile up; assemblies don't. The archive holds 42.7M read sets but only 5.1M assemblies.
12 / 100assemblies per 100 read sets in ENA
2
Multi-tool taxonomic profilingtaxprofilerEffort concentrates on a few pathogens; a second classifier lets a facility cross-check the call.
36,565E. coli genomes — 15.4% of the top-50 cohort
3
Amplicon analysisampliseq16S/18S/ITS is a whole domain the metagenomics-heavy shelf doesn't serve yet.
0 of 7current pipelines handle amplicon

Why this order. Assembly first — the gap is widest. Profiling second — it deepens what the shelf is already strong at. Amplicon third — it opens a new domain. The order is a judgement, stated as one.

Bacterial assembly first

Reads pile up; assemblies don’t. The public archive holds 42,674,183 read sets but only 5,117,626 assemblies — about 12 per 100. A streamlined bacterial-assembly workflow turns deposited reads into the genomes people actually compare.

Then multi-tool profiling

Effort piles onto a few pathogens: Escherichia coli alone is 36,565 genomes and 983,438 papers — 15.4% of the top-50 cohort. A multi-classifier profiler lets a facility confirm a call with more than one tool instead of trusting one.

Then amplicon

Not every study is shotgun. 16S/18S/ITS work is a whole domain the metagenomics-heavy shelf doesn’t serve — 0 of 7 current pipelines handle it. Ampliseq adds ASV tables, taxonomy, and diversity metrics, in the same place as everything else.

Why this order

Assembly first because the gap is widest and most concrete. Profiling second because it deepens what the shelf is already strong at. Amplicon third because it opens a new domain rather than improving one. The order is a judgement, and we state it as one.

Beta is the on-ramp

Kraken2 + Bracken already ships in beta as the lightweight per-sample profiler; the planned multi-tool path builds on that demand, it doesn’t replace it. The roadmap moves things along a track — planned → beta → available — not in one leap.

New pipelines are the visible part of a roadmap. The higher-return work is quieter: making the outputs we already produce easier to read, link, and trust. None of it needs a new workflow.

Making outputs usableno new pipeline required

these add no new pipeline, touch data that never leaves the server, and make every current run more useful

Fill in the metadata you already collected56%
Make every report previewable in place2 of 4
Link every result to the data behind it30%
Carry QC forward instead of re-running it1 report

Fill in the metadata you already collected

Even in its best year only 74% of new ENA samples carried a real collection date; by 2025 that had fallen to 56%, and location is filled for just 58.5% of the year’s 4.95M samples. SeqDesk captures MIxS at intake — surfacing and validating those fields on every output is cheap and closes the gap directly.

Make every report previewable in place

Study MultiQC and NanoPlot already emit previewable HTML; the MAG path and future runs should render the same way. An output you can open without downloading is one people actually read — no new workflow, just consistent rendering.

Link every result to the data behind it

In 2025 only 51% of open-access papers linked any underlying dataset, and just 30% of DataCite’s 129.8M-DOI corpus carries a machine-readable licence. SeqDesk holds the run, the sample, and the accession together, so emitting that link and a licence on every result is low-effort — and rare in the wild.

Carry QC forward instead of re-running it

Study MultiQC already aggregates a study's existing QC into one report rather than recomputing it. Extending that pattern across pipelines makes each new run cheaper and each study easier to read as a whole.

Why fruit beats features

These four add no new pipeline, touch data that never leaves the server, and make every current run more useful. They are the highest-return work on the board — which is why they sit between what's next and the whole.

All of it points at one position. A pipeline runner is useful; a tool that takes a sample from intake, through analysis, to a citable accession — without the data ever leaving — is rare.

The integrated wholeof 19 tools compared

run pipelines8

order intake5

ENA submission4

all three1

1 / 19does order intake, pipelines, and ENA submission in one place

3 → ~67 PbpENA public raw bases, 2016 → 2025 — ~2× ahead of the 32-month-doubling forecast (~31 Pbp)

4 run today · 3 planned · fruit throughout. Not a finished platform — a track, and where we stand on it.

Pipelines are one third of the job

Of 19 tools compared for sequencing facilities, 8 run pipelines, 5 do order intake, and 4 submit to ENA. Running a workflow is necessary, not sufficient — a facility needs all three to take a sample from intake to a citable accession.

One tool does all three

Just 1 of the 19 natively does order intake, pipelines, and ENA submission in one place. That is the position the roadmap defends: every new pipeline lands inside the same app that already took the order and can submit the result.

The deluge is the reason to integrate

ENA’s public raw sequence data grew from 3 to ~67 Pbp between 2016 and 2025 — about 2× ahead of its own 32-month-doubling forecast, which predicted ~31 Pbp. Volume at that scale rewards a facility that can run, describe, and submit without moving data between tools.

Where we are, and what's next

Today: four pipelines you can run, one of them production. Next: assembly, multi-tool profiling, amplicon. Throughout: filling metadata, previewing reports, linking data. We are not promising a finished platform — we are showing the track and where on it we stand.

That is the whole point of doing this inside one application. You can browse the pipeline store, read how pipelines run inside SeqDesk, or see how a result climbs the wire to ENA.

Sources. The pipeline shelf and its statuses are read directly from this site’s pipeline store data, which mirrors the live store in the main SeqDesk app; the production MAG workflow lives in the metaxpath repository. The “why” numbers come from the /data analyses on this site: assembly rate, taxon popularity, metadata completeness, open data, and the ENA doubling. The integration counts (8 / 5 / 4 across 19 tools, 1 doing all three) are from Comparing tools for sequencing facilities. Shelf counts are derived in code from the store data, not hand-entered, so they can’t drift from what the site lists.