gyrB (bacterial DNA gyrase subunit B)

gyrB encodes the B subunit of DNA gyrase, a protein-coding single-copy gene in bacteria. Because it is protein-coding, it accumulates substitutions faster than the 16S rRNA gene, so it separates organisms inside a genus that a 16S V3–V4 amplicon leaves merged. The cost is a reference database much smaller than SILVA, and a marker that is bacterial rather than universal.

MetaFlux runs gyrB through the same amplicon path as every other marker — PhiX removal, primer trimming, DADA2 denoising, length filtering, taxonomy. Three things set it apart from the rRNA markers, all of them consequences of its reference database (workflow/markers/gyrB.yaml declares them): the reference is already amplicon-length, only the RDP classifier is available, and the reference carries a built-in paralog tag. The first needs no attention at all; the other two need one config line each.

At a glance

amplicon.type	gyrB (also accepted as gyrb or GYRB — matched case-insensitively)
Reference database	DD7RZ8 v6, the INRAE gyrB DADA2 trainset built on GTDB r226
Role of that database	both the taxonomy reference and the amplicon-length probe substrate
Probe mode	direct — lengths read straight off the reference
Classifier	RDP only (amplicon.taxonomy.method: rdp)
Separate species step	none — the trainset itself carries species
Region extractor	none (no gyrB equivalent of Metaxa2 or ITSx)
Rank model	Gene;Phylum;Class;Order;Family;Genus;Species
Primers	none are shipped; Barret 2015 F64 / R353 is the pair the reference was built with

Setting up a run

Set the marker type, point at the primer FASTAs, and choose the RDP classifier. A minimal gyrB block looks like this:

mode: amplicon

amplicon:
  type: gyrB
  primers:
    fwd: /path/to/gyrB_F64.fasta
    rev: /path/to/gyrB_R353.fasta
    orientation: fixed          # fixed | mixed
  expected_length: auto
  probe_length_stat:
    16S: p95                    # gyrB reads this key — see "Amplicon length" below
  extraction:
    enabled: false              # no gyrB region extractor
  taxonomy:
    method: rdp                 # gyrB has no SINTAX build
    min_boot: 80
    try_rc: true
    filter:
      enabled: true
      keep:    []
      discard: []               # [other] drops paralogs — see "The paralog tag"

Primers are never supplied by MetaFlux: amplicon.primers.fwd and amplicon.primers.rev must both point at existing FASTA files, and parsing stops immediately if either is missing. The Barret 2015 F64 / R353 pair is the natural choice, because DD7RZ8's reference amplicons were cut with those same primers, but any gyrB pair works — the length window is measured from the reference rather than assumed.

The shipped config template defaults to method: sintax

config/config.yaml ships with amplicon.taxonomy.method: sintax, which is fine for 16S/ITS/18S but not available for gyrB. A run config copied from the template without changing that line stops at parse time, before any compute is spent:

[MetaFlux] marker gyrB has no SINTAX reference database, so
amplicon.taxonomy.method: sintax is not available for it. Use method: rdp.

The reference database

DD7RZ8 v6 is a gyrB trainset in DADA2 format, built at INRAE on GTDB release r226. It is downloaded on first use and cached, so later runs reuse it.

Cached at	refdb/gyrb/train_set_gyrB_v6.fa.gz
Source	Recherche Data Gouv (INRAE Dataverse) file-access endpoint, https://entrepot.recherche.data.gouv.fr/api/access/datafile/760120
Fetch rule	fetch_gyrb_dada2, generated from the URL in the marker pack
Log	<out_dir>/logs/refdb/fetch_gyrb_dada2.log

The download is a single plain .fa.gz file, so no rule was written by hand for it: MetaFlux generates one fetch rule per plain-download reference declared by the active marker pack. The URL is a Dataverse API endpoint that redirects to the object store, which wget follows.

The cache root is references.refdb_root (default refdb). An entry under references is only needed to relocate a database that already exists elsewhere on disk — for gyrB that entry is:

references:
  gyrb:
    dada2: /shared/dbs/train_set_gyrB_v6.fa.gz

Every reference sequence is one primer-trimmed in-silico amplicon, not a full-length gene, and the header is a GTDB lineage whose ranks already carry their prefixes, preceded by a bare gene tag:

>{gene};p__Phylum;c__Class;o__Order;f__Family;g__Genus;s__Genus_species

Because the prefixes are already in the reference, the marker pack sets prefix_style: embedded and MetaFlux emits each rank value verbatim instead of prepending its own p__/c__/… — the species slot stays a GTDB binomial such as s__Actinotalea_bogoriensis.

Amplicon length: measured directly, not by in-silico PCR

For 16S, 18S and rpoB the amplicon-length probe runs an in-silico PCR: two passes of cutadapt against a full-length reference, keeping the sequences that carry the forward primer at the 5′ end and the reverse complement of the reverse primer at the 3′ end. That is impossible against DD7RZ8, whose sequences have already had those primer sites cut off upstream — the PCR would find nothing to match and return almost no amplicons.

So the gyrB pack sets probe_mode: direct: cutadapt is not run at all, and the length distribution is read straight from the reference sequences. Measured across the shipped DD7RZ8 v6 file, that distribution is tight — median 247 bp, minimum 234 bp, maximum 270 bp — which is exactly what a pre-trimmed amplicon reference should look like.

The probe only runs when amplicon.expected_length: auto. It writes its statistics to a JSON file whose name encodes the marker, the reference tag and a 12-character hash of both primer FASTAs:

refdb/cache/probe_gyrB_gyrb_dd7rz8_v6_<primer_hash>.json

Changing primers changes the hash, so the probe reruns; keeping the same primers across runs reuses the cached result. That JSON then feeds two downstream steps: pick_trunclen, which uses the resolved expected length to enforce the forward/reverse overlap constraint on truncLen, and the ASV length filter, which builds its auto window from the probe's q1 and p95 plus length_filter.window_margin. See Amplicon length and truncation for both.

gyrB reads probe_length_stat.16S, not a key of its own

Which statistic of the probe distribution becomes expected_length is chosen by amplicon.probe_length_stat.<key>, and the gyrB pack deliberately reuses the 16S key rather than defining a separate one. If the 16S entry is missing, parsing stops with:

[MetaFlux] amplicon.probe_length_stat is missing the '16S' key that marker gyrB
needs. Add it, e.g.
    probe_length_stat:
      16S: p95

The key is checked at parse time for every non-ITS marker, whether or not expected_length is auto. A manual expected_length (a single integer, or a [min, max] pair) skips the probe itself, but the 16S entry must still be present.

Taxonomy: RDP only

The DD7RZ8 release ships a DADA2 trainset and nothing else — there is no SINTAX-formatted build of it — so the marker pack leaves taxonomy_sintax_db empty and MetaFlux refuses method: sintax at parse time rather than failing later inside the classifier.

Classification is DADA2's assignTaxonomy (the RDP naive Bayesian classifier) against the trainset, with taxonomy.min_boot as the bootstrap confidence minimum and taxonomy.try_rc deciding whether reverse-complemented sequences are also tried. There is no separate addSpecies step: unlike 16S, where species come from a second SILVA file, the gyrB trainset is itself labelled to species, so the classifier reaches the species rank directly. amplicon.seed fixes the bootstrap RNG, so RDP results are reproducible run to run.

The classifier writes seven rank columns, taken from the marker pack:

Slot	Rank column	Prefix in the taxonomy string
1	Gene	none — a bare gyrB or other tag
2	Phylum	p__
3	Class	c__
4	Order	o__
5	Family	f__
6	Genus	g__
7	Species	s__

The paralog tag

DNA gyrase subunit B has a close relative, topoisomerase IV subunit B (parE), similar enough that gyrB primers co-amplify it. DD7RZ8 handles this by labelling every reference sequence in its first rank slot: gyrB for the true gene, other for a co-amplified paralog such as parE.

MetaFlux keeps that tag as an ordinary taxonomy rank, which means paralogs can be removed with the same keep/discard filter used to drop chloroplast and mitochondrial ASVs from a 16S run — no extra machinery, no separate step:

amplicon:
  taxonomy:
    filter:
      enabled: true
      discard: [other]     # not the same as keep: [gyrB] — see below

Filter tokens are matched against whole segments of the taxonomy string, so other matches the gene tag and only the gene tag — never a substring somewhere inside a species name. Because the tag carries no rank prefix, the token is written bare, unlike the prefixed tokens (k__Bacteria, o__Chloroplast) used for the rRNA markers. See Keeping and discarding taxa.

discard: [other] drops the paralogs, and keep: [gyrB] is not an equivalent way of writing it. An ASV the classifier could not tag at all has an empty taxonomy string, so it carries no segment for either list to match: discard leaves it in, keep throws it out. In the shipped gyrB test run that is 5 of 256 ASVs — 2% lost silently by the keep form.

Leave the filter empty on the first run

How much other to expect depends on the primers and the samples, so there is no universal right answer and MetaFlux applies no default. Running once with empty keep/discard lists makes the paralog fraction visible in the output — count the ASVs whose taxonomy string starts with other, or read the Gene column of 6.taxonomy/taxon_seq_table.txt — and then decide whether to discard them.

The fraction is not random noise: co-amplification of parE is taxon-dependent, so it concentrates in particular clades rather than spreading evenly across the table. A run dominated by the affected groups will show a much larger other fraction than one that is not, which is worth checking against the composition before assuming a primer problem.

If a keep/discard combination removes every ASV, assign_taxonomy stops with an error instead of writing a header-only table, since that is nearly always a marker/config mismatch.

GTDB splits genera, and the filter matches whole segments exactly

DD7RZ8 v6 carries GTDB r226 taxonomy, and GTDB routinely divides a genus that is polyphyletic under NCBI into several suffixed genera — g__Pseudomonas_A, g__Pseudomonas_B, and so on. In the shipped reference, 734 of 5,207 genera carry such a suffix, and Pseudomonas alone appears as seven separate names:

g__Pseudomonas   g__Pseudomonas_B   g__Pseudomonas_E   g__Pseudomonas_M
g__Pseudomonas_O   g__Pseudomonas_P   g__Pseudomonas_T

The keep/discard filter matches whole ;-delimited segments exactly — deliberately, so that g__Bacillus can never match a substring elsewhere in a lineage. The consequence here is that discard: [g__Pseudomonas] removes only the unsuffixed genus and leaves the other six untouched, and keep: [g__Pseudomonas] keeps only that one.

Check the names actually present before writing a filter — the Genus column of 6.taxonomy/taxon_seq_table.txt shows them — and list every variant that applies. This affects gyrB only: rpoB's FROGS reference carries NCBI taxonomy, where the genus is not split.

No region extractor

MetaFlux has no gyrB equivalent of Metaxa2 (16S) or ITSx (ITS), so there is nothing to trim ASVs down to a target region. amplicon.extraction.enabled: true is therefore forced off for gyrB with a warning on stderr:

[MetaFlux] warning: marker gyrB has no target-region extractor; ignoring
amplicon.extraction.enabled: true (no extraction step will run)

Setting enabled: false explicitly keeps the run log clean. The ASV length filter still runs either way, so off-target amplicons of the wrong size are still removed.

What the output looks like

The final tables are the same three files as for every other marker, under 6.taxonomy/ in the output directory:

File	Rows	Notes
asv_table.txt	ASV IDs	per-sample counts plus one taxonomy column
asv_table_seqs.txt	ASV sequences	the same table keyed by sequence instead of ID
taxon_seq_table.txt	ASV IDs	one column per rank (Gene … Species) plus the sequence

A gyrB taxonomy string keeps the gene tag in front of the GTDB lineage:

gyrB;p__Pseudomonadota;c__Gammaproteobacteria;o__Pseudomonadales;f__Pseudomonadaceae;g__Pseudomonas;s__Pseudomonas_aeruginosa

Ranks the classifier could not assign above min_boot are simply absent from the string, so a shorter string means a less confident assignment rather than a missing rank.

Related pages

• rpoB — the other protein-coding marker, which differs in almost every detail listed above except being RDP-only
• Amplicon length and truncation
• Keeping and discarding taxa
• Configuration

References

20. Briand, M., Rué, O. & Barret, M. (2025). gyrB database for taxonomic assignment formatted for DADA2 (train_set_gyrB_v6). Recherche Data Gouv (INRAE Dataverse). https://doi.org/10.57745/DD7RZ8

22. Parks, D. H., et al. (2022). GTDB: an ongoing census of bacterial and archaeal diversity through a phylogenetically consistent, rank normalized and complete genome-based taxonomy. Nucleic Acids Research. (Genome Taxonomy Database, release r226)

31. Barret, M., et al. (2015). Emergence shapes the structure of the seed microbiota. Applied and Environmental Microbiology. (gyrB primers F64 / R353; also used to build the DD7RZ8 reference amplicons)

The numbering follows the full list on Citation and references.