FILE HG–2026–08

ENGINE phasing-based closure

CLASS. TECHNICAL · PUBLIC

TRACE 2026–08–04

SCOPE PURE ISOLATE

05 Ingestion · whole genome

HoloGen

It closes genomes where the reference standard states in writing that it does not apply: without reads long enough to span the repeats. There, HoloGen turns depth into resolution and delivers the genome with its certainty measured.

This is the regime of the food and environmental sample: fragmented DNA, scarce biomass, short libraries. The conventional workflow returns a broken draft or nothing; here the data is recovered instead of lost. And the assembly does not end in a file: it feeds target discovery and risk interpretation without leaving the system.

How to read this page

The figures on this page come from a single trace of the engine running end to end. What is not in the trace is not here.

Pediococcus acidilactici · ONT
2.462,7 s wall-clock · 2026-08-04

1.997.608 pb

chromosome closed into a circle, with its four junctions observed against the reads

208×

of depth over the emitted genome, with a read N50 of only 2,685 bp

independent lines of evidence: none is believed on its own

4,28

copies measured where the assembler annotated 162. Annotation does not decide; measurement does

What makes it unique

It is the only engine that delivers a closed genome in the regime the gold standard rules out in writing: without reads able to span the repeat. Instead of demanding length, it spends depth — variant phasing, multiplicity measured against the reads, real consensus and validation of every seam — and recovers the data the conventional workflow loses entirely.

With the same discipline in the other direction: multiplicity comes from counting k-mers in the reads, not from the assembler's annotation — 162 annotated against 4.28 measured, 10.8 annotated against 15.17 measured — and a figure that does not fall close to an integer is declared unresolved instead of being silently rounded. Certainty measured, not promised.

01 The state of the art

Eight tools, eight formats, and a file at the end.

Chaining assembler, quality control, annotator and viewer by hand means re-exporting at every hop and losing traceability between steps. And when the assembly does not close, most workflows give up without saying why.

Tool

How far it goes

What HoloGen adds

Trycycler · the manual gold standard

Consensus assembly: it combines several independent drafts and produces the most accurate genome published to date when conditions allow.

Closes the genome in the very territory Trycycler declares outside its scope — depth below 25× or reads too short for complete assembly — with no human intervention and a deterministic output: same data, same genome.

Autocycler, Hybracter, Dragonflye

They automate consensus and polishing best practice with no manual intervention, at a scale of hundreds of genomes.

Works on the case those pipelines refer to manual curation: it measures the multiplicity of the large repeat against the reads and delivers the result with its certainty stated.

Flye, SPAdes, Unicycler on their own

They produce the draft and its graph, which is where the work begins.

Checks the graph against the reads before handing it over: where the annotation said multiplicity 162, the coverage-based measurement gives 4.28 — and that figure travels with the genome.

A hand-built tool chain

Covers quality, assembly, annotation and visualisation if the laboratory has a dedicated bioinformatician.

Chains quality, assembly, annotation and visualisation into a single traceable step, with an intermediate verdict: if it is not worth going on, it says so before the hours are spent.

Turnkey sequencing service

Returns the assembly and a standard report without the laboratory touching anything.

Returns the genome inside the workflow: it feeds straight into targets, risk and comparison with what you already have — and with difficult material it keeps measuring instead of stopping at the draft.

02 The closure engine

Where the standard withdraws, the work begins here.

No read in the measured case spans the 40,256 bp element: the longest is 50,005 bp and it would have to cover the whole repeat plus two kilobases of anchoring on each side. That is where the gold standard declares itself inapplicable, and where 208× of depth was going unused.

1.997.608 pb

chromosome closed into a circle, with its four junctions observed against the reads. Not inferred from the graph: observed.

And what is left out is declared just as plainly: a 5,164 bp repeat at 4.28 doses that the graph cannot place. A closure that does not say what it left out is not a closure, it is an assertion.

Line 01

Multiplicity measured against reads

The copy count of each segment comes from sampling k-mers inside it and counting how often they appear in the reads, not from the assembler's annotation.

Line 02

Variant phasing

The differences between copies of the same repeat allow them to be separated when no read spans it whole. This is the line that turns depth into resolution.

Line 03

Real consensus over the reads

The emitted sequence is rebuilt from the reads that support it; it is not inherited from the draft.

Line 04

Seam gate

Every proposed junction must be observed in reads that cross it with anchoring on both sides. Those that fail do not get in.

What is NOT claimed from these figures

They come from one case measured end to end, not from a cross-validation campaign against the tools cited. What they demonstrate is that the genome closes in a regime where the gold standard declares itself inapplicable; what they do not authorise is any claim about how many accuracy points separate one from the other where both do apply.

And the engine works on filtered reads, so its measured haploid depth is 116× where the measurement on the raw file gives 141×. Both are correct; the engine's figure is published, and which is which is stated so that nobody compares apples with oranges.

03 How it works

From the read file to the navigable genome.

Every long step emits its explicit verdict and its real progress, so that a forty-minute process is not a spinner with no information.

Quality control and read filtering

Filtering by length and quality with metrics before and after. The engine records that it works on filtered reads, because that shifts every depth figure it publishes afterwards.

Assembly by platform

SPAdes for short reads, Flye for long reads, chosen from the detected platform. Automatic quality control with BUSCO and contiguity metrics.

Phasing-based closure

If the draft does not close, the closure engine steps in: it measures multiplicities, phases variants, rebuilds consensus and validates every seam. It always delivers the best result the data supports, with its status stated.

Species identification

Kraken2 and ribosomal gene extraction with barrnap, plus alignment against a canonical reference, and name resolution against the curated knowledge taxonomy. If that layer is unavailable, it returns the raw name instead of failing.

Annotation

Prokka or Bakta, predicting coding genes, ribosomal RNA and transfer RNA.

Characterisation and navigation

Resistance determinants, virulence factors, plasmids and mobile elements, on a genome browser where every finding is seen in its position.

04 The contract

What goes in, what comes out, what it chains to.

Reads from a pure isolate: paired-end Illumina or nanopore, including the chunked delivery the sequencer itself produces during the run. Mixed samples go to OmniOta.

Out

Assembly with its quality control, species resolved against canonical taxonomy, full annotation, characterisation of resistance, virulence, plasmids and mobile elements, and a genome browser.

Chains to

Direct output to BioMiner for target discovery and to GIFCloud for risk interpretation and epidemiological tracing. An assembly that is not used downstream is a dead file.

05 Technical sheet

Name

Holistic Genome Assembly Engine: an assembly and genome-closure engine for isolates.

Input

Reads from a pure isolate. Paired-end Illumina or nanopore at a minimum quality of Q10, including chunked upload during the sequencer run. The upload survives navigating away from the page.

Assemblers

SPAdes for short reads and Flye for long ones, with selection assisted by the platform detected in the file itself.

Closure

A variant-phasing engine with four independent lines of evidence, real consensus and a seam validation gate. Multiplicity measured against the reads, never read off the graph.

Long steps

Every step emits an explicit verdict and real progress. A forty-minute process reports where it is, not merely that it is still alive.

Reassembly

History per sample: you can reassemble with different parameters and compare against the previous run without losing it.

Limit · scope

Epidemiological typing — MLST, cgMLST and wgMLST — leaves the pipeline: it depends on schemes that exist for only a minority of organisms and its interpretation is GIFCloud's territory. HoloGen assembles and closes.

Limit · sample

The engine assumes a pure isolate. A mixed sample is not assembled here: it goes to OmniOta, the module written for communities.

Limit · figures

The published closure figures come from one case measured end to end, not from a cross-validation campaign against Trycycler, Autocycler or Hybracter. They show how the engine behaves where those tools declare themselves inapplicable, not a quantified advantage where they do apply.

Assembly: SPAdes and Flye. Quality control: BUSCO and contiguity metrics. Species identification: Kraken2, barrnap and minimap2 against a canonical reference. Annotation: Prokka and Bakta. Read filtering: filtlong. Genome browser: IGV.js. Closure engine trace: 2026-08-04, 2,462.7 s.
Declared scope of the gold standard: Wick, Judd and Holt (2021) Genome Biology 22:266 · 10.1186/s13059-021-02483-z — "if complete assemblies are not possible (e.g. due to insufficient read length) or read depth is shallow, then Trycycler is not appropriate". Autocycler: Wick et al. (2025) bioRxiv 2025.05.12.653612. Hybracter: Bouras et al. (2023) bioRxiv 2023.12.12.571215.

Always deliver what the data supports. And say what is left out.

Request access See the twelve modules

BIOTECNO.org · Vitoria-Gasteiz EU + Codex