Nanopore mode¶

mode: nanopore — Oxford Nanopore Technologies (ONT) long reads in, a reoriented, polished Flye assembly out. Seven rules, in workflow/rules/nanopore/.

flowchart TD
    subgraph FE["mode: nanopore, rules in workflow/rules/nanopore/"]
        direction TB
        RAW([ONT reads as basecalled])
        QC1[raw_long_read_qc<br/>NanoPlot, before]
        FL[filter_long_reads<br/>Filtlong]
        QC2[filtered_long_read_qc<br/>NanoPlot, after]
        ASM[ont_assembly<br/>Flye]
        FIX[fix_start<br/>dnaapler, reorient to dnaA]
        POL[long_read_consensus<br/>Medaka]
        FIN[finalize_contigs]

        RAW --> QC1
        RAW --> FL --> QC2
        FL --> ASM --> FIX --> POL --> FIN
    end

    subgraph MC["checked before the assembler starts"]
        direction TB
        MK[check_medaka_model<br/>resolves the Medaka model]
    end
    FL --> MK
    MK --> POL

    subgraph SH["shared, workflow/rules/shared/"]
        direction TB
        SC[contamination screen<br/>BLAST, BlobTools, genus selector]
        OUT([contigs_final.fasta])
        SC --> OUT
    end
    FIN --> SC

There is no PhiX step: PhiX is an Illumina spike-in and does not exist in an ONT library. The Medaka model is resolved before Flye runs, so a bad model name costs seconds rather than an hour of assembly — see the model-choice note.

Terms used on this page


CARD	Comprehensive Antibiotic Resistance Database

1. Read filtering, and why it is the biggest lever in this mode¶

An ONT run produces reads of wildly varying length and quality, and far more data than a bacterial isolate needs. Assembling everything is slower and worse: short, poor reads add errors without adding contiguity. filter_long_reads runs Filtlong with three settings:

Flag	From	Default	What it does
`--min_length`	`parameters.long_read_qc.min_length`	1000	drops reads too short to span a repeat, which is ONT's whole advantage
`--keep_percent`	`parameters.long_read_qc.keep_percent`	95	keeps the best N% of the remaining bases, by Filtlong's own score
`--target_bases`	fixed at 5×10⁸	—	stop at ~100× for a 5 Mbp genome; more costs Flye time and buys nothing

keep_percent is what decides whether a small plasmid survives

Filtlong ranks reads by a blend of length and quality, and keep_percent discards the bottom of that ranking. A small plasmid cannot produce reads longer than itself, so its reads sit near the bottom by construction and go first.

Measured on K. pneumoniae TUM24772, whose 5,596 bp Col2 plasmid was missing from the assembly while sitting complete in the Illumina data: of the 603 raw ONT reads covering it, 93 survived the old keep_percent: 90, and 500 survive the 95 that now ships. Recovering the reads was necessary but not sufficient — no setting tried, raw unfiltered reads included, assembled that plasmid.

parameters.long_read_qc.length_weight has no effect in this mode: this rule passes no --length_weight, so Filtlong uses its own default of 1. That key reaches Filtlong in hybrid mode only.

NanoPlot runs on both sides of the filter — 01.reads/{sample}/ont/raw_qc/ and filt_qc/ — so what the filter removed is visible rather than assumed. Read the two side by side.

2. The Medaka model is resolved before anything is assembled¶

check_medaka_model (a shared rule) runs straight after read filtering, validates parameters.nanopore.medaka_model or infers it from the basecaller tag in the read headers, and writes the resolved name to {sample}_medaka_model.txt. The assembler waits on that file. A typo therefore fails in seconds with a table of suggestions, instead of an hour later when Flye has already finished.

`medaka_model`	Effect
`auto` (default)	infer from the FASTQ headers; fails with suggestions if they carry no tag
an explicit name	validated against the installed model list
`false`	Medaka is skipped, and the rule is not defined at all

Which model auto resolves to, why it is the bacterial variant, and when to pin a different one, are in The Medaka polishing model.

Setting the model also changes the assembler

With parameters.nanopore.flye_input_mode: auto (the default), Flye's read mode is derived from the Medaka model name: a name containing fast selects --nano-raw, anything else --nano-hq. Fast basecalling means noisier reads, so the coupling is deliberate — but it means a change of polisher model silently changes the assembly. Set flye_input_mode explicitly to pin one without the other. The reasoning, and the cases where inference fails, are in methods_medaka_model_choice.md.

3. Assembly and reorientation¶

ont_assembly runs Flye with --iterations 5 of its own internal long-read polishing. Long reads span the repeats that break a short-read assembly, so Flye usually closes a bacterial chromosome into one circular contig.

A circular sequence has no natural start, and Flye's arbitrary one makes two assemblies of the same isolate look different. fix_start runs dnaapler all, which searches every contig for the canonical start genes at once — dnaA for a chromosome, repA for a plasmid, terL for a phage — and rotates the sequence to begin there. It runs at a fixed seed, so a re-run is reproducible.

Two by-products are used later:

ignore_list.txt, written from Flye's assembly_info.txt, lists the contigs Flye did not call circular. Rotating a linear contig would move its true ends into the middle, so dnaapler is told to leave those alone.
{sample}_all_reorientation_summary.tsv records which start gene was found on each contig and how convincingly. Together with Flye's circularity column it builds the --replicons table handed to Bakta, so the annotation knows which sequences are closed chromosomes and can call a gene that runs across the origin.

Headers are trimmed to their first whitespace token here, because Bakta, Platon, geNomad and the BLAST screen all join on that token.

4. Order: screen first, polish second¶

The contamination screen runs on the reoriented assembly, and Medaka polishes the decontaminated one. Polishing a contaminant contig with this isolate's reads would waste effort and could smear real differences.

finalize_contigs then copies whichever file ended the chain — the Medaka consensus, or the decontaminated assembly when Medaka is off — to contigs_final.fasta, and logs which one it copied. That one line is the provenance of the delivered genome.

What this mode has that the short-read modes do not¶

Topology. Flye's circularity calls reach Bakta, so closed replicons are annotated as closed.
Contiguity. The mobilome module's element boundaries are far more meaningful on a closed genome; on a draft the class survives fragmentation but the extent does not (Draft assemblies).

What it gives up¶

No read-based CARD screen. That leg needs short reads. ABRicate on the contigs still runs (Antimicrobial resistance).
No length or coverage filter on the contigs. The reoriented assembly reaches the screen as Flye left it.
Residual base errors are what Medaka fixes; switching it off leaves them in. Indels in homopolymers frameshift genes and ruin an annotation.

What it writes¶

Path	Contents
`01.reads/{sample}/ont/raw_qc/`, `filt_qc/`	NanoPlot, before and after filtering
`01.reads/{sample}/ont/{sample}_filt.fastq`	the reads Flye assembled and Medaka polished
`02.assembly/{sample}/flye/`	`assembly.fasta`, `assembly_info.txt`, `ignore_list.txt`
`02.assembly/{sample}/fix_start/`	dnaapler's output and the reorientation summary
`02.assembly/{sample}/medaka/`	`consensus.fasta` (absent when Medaka is off)
`02.assembly/{sample}/{sample}_replicons.tsv` + `_audit`	the table Bakta is given, and the reason per contig
`02.assembly/{sample}/contigs_final.fasta`	the delivered genome