Skip to content

Actinorhodin: gene order, annotations and genomic distance

Question: In the known actinorhodin biosynthetic region of Streptomyces coelicolor A3(2), do the annotated synthase, chain-length factor and acyl carrier protein occur in the expected order, and how many genomic bases separate them?

The original genome publication, Figure 3 identifies the actinorhodin cluster as SCO5071–SCO5092. The retained RefSeq annotation snapshot maps those old locus tags to current gene IDs on the linear chromosome NC_003888.3. This exercises a real BGC without making an NMDC annotation convention part of the general model.

Independently recorded expectations

These values were read directly from the retained GFF rows before running the converter or SQL. expected.json is maintained evidence, not generated query output. report.json checks the result against it and records full-source line numbers and coordinates for all 22 genes, with a separate direct source check of the full order and the three focus intervals. The three focus CDS rows are independently inspected for parent gene identity, CDS identity, product, strand and coordinates; their source line numbers and values are recorded in source_cds_evidence.

Old locus tag Current gene CDS product in this snapshot Inclusive genomic interval Strand
SCO5087 SC_RS27595 beta-ketoacyl-[acyl-carrier-protein] synthase family protein 5,529,801–5,531,204 +
SCO5088 SC_RS27600 ketosynthase chain-length factor 5,531,201–5,532,424 +
SCO5089 SC_RS27605 acyl carrier protein 5,532,449–5,532,709 +

Ordering uses increasing chromosome coordinates, irrespective of transcriptional strand. Signed gap is next.start - previous.end - 1: −4 means a four-base overlap; 24 means 24 intervening bases. These are nucleotide distances, never amino-acid offsets. The query selects these three old locus tags explicitly; a general overlap query at 5,532,706 also finds the next gene, SCO5090.

The full cluster query recovers all 22 locus tags in order. Exact product lookup finds SCO5089 for acyl carrier protein. Controls give no matches for an absent product, an interval with no selected features or a different sequence accession version. These overlap queries allow positive intervals outside the selected region, including beyond the reference sequence; no matching rows means an empty result. This is not a reference-length validation API. Queries declaring a different reference context or protein coordinate space fail explicitly. SQL parameters preserve literal attribute values. The reference guard derives refseq:ACCESSION.VERSION from the sole chromosome stored in DuckDB. It cannot be bypassed by supplying two matching caller claims. This is a bounded single-RefSeq-sequence query, not a multi-assembly join API.

Source, selection and limits

The full source is the unchanged 4.69 MB NCBI annotwriter response retrieved September 21, 2026. The download endpoint returned the whole chromosome despite range parameters; the index records that exact request, checksum and outcome. Sequence accession version identifies the reference; the retrieval date and SHA-256 pin the annotation snapshot, because annotations can change without changing sequence version. Reuse follows the NCBI data policy.

The derived input contains the 22 selected gene rows and their 22 direct CDS children, original header, unchanged row bytes and absolute chromosome coordinates, plus explicit selection provenance. Selection is by the publication's locus tags, not inferred biological exclusion. The complete source remains available to reproduce and inspect the selection. Both files pass the pinned GenomeTools validator.

gff3-contig/4.0.0 converts this declared excerpt with reference context refseq:NC_003888.3. Exact export recovers the excerpt byte for byte. Reconstruction derives its feature rows from modeled fields and semantic mappings, then reimports with identical fields and relationships. This does not claim that 44 selected rows reconstruct the complete chromosome annotation. Edited bundles are rejected by both exporters when mapped fields or preservation records change. The exercise's export_example also compares the declared reference, source URI and metadata profile with its independently pinned context. The generic conversion CLI treats those declarations as caller-supplied provenance; source bytes alone cannot authenticate them. bgc-check verifies both source and pinned context.

The query describes annotations and proximity. It does not infer biochemical function from distance or establish cluster activity. MIBiG provides complementary curated evidence on accession AL645882.2; its wider display region and that accession are not silently substituted for this exercise's RefSeq coordinates.

Reproduce offline

After installing the repository's pinned conversion dependencies:

just bgc-check    # read-only selection, conversion, query and report verification
just bgc-report   # explicitly regenerate the derived excerpt and measured report
mkdir -p local/bgc-demo
just conversion-import corpus/derived-examples/actinorhodin.gff3 \
  gff3-contig/4.0.0 refseq:NC_003888.3 local/bgc-demo/bundle.json --metadata-profile ncbi
just conversion-export local/bgc-demo/bundle.json local/bgc-demo/exact.gff3 exact
just conversion-export local/bgc-demo/bundle.json local/bgc-demo/reconstructed.gff3 reconstruct

The example builds a temporary DuckDB in ignored local/, validates the model, executes genomic gene/CDS/attribute joins, and removes the temporary database. Set UV_OFFLINE=1 to require cached dependencies. just check and CI include the read-only check and regression tests. The general conversion report measures this case too.