id: nmdc-pfam-protein/5.0.0
format: gff3
status: experimental
scope: NMDC HMMER Pfam protein-relative annotation rows with an explicit CDS/FASTA context
required_context:
  - source-scoped or assembly-qualified reference context
  - closed Dataset containing the referenced contig-coordinate CDSs and their translated sequences
  - unique protein-reference to CDS bindings and two distinct artifacts with structural_annotation/protein_sequence roles, URIs and SHA-256 provenance
mapping:
  coordinates: 1-based inclusive amino-acid offsets within the bound CDS translation
  seqid: source column 1 is a protein reference; model seqid is the bound CDS, whose own seqid is its contig (issue 40; 2.0.0 used the contig)
  parent: model parent is the explicit CDS binding; no source Parent tag is invented
  type: literal Pfam accession, not a Sequence Ontology assertion
  score_type: bit_score on every row with a score (HMMER column 6, per nmdc-lakehouse docs/pfam_annotation_gff.md); export never reads it
  translation_table: not derived; the protein context Dataset is used as supplied
  target: target_id is the Pfam accession, target_start and target_end come from model_start and model_end (from 5.0.0, issue 94); the attributes stay and export refuses a target that disagrees with them
  attributes: ordered generic pairs with repeated/empty/unknown values and comma grouping retained
  typed_attributes: Name -> name, Note -> note, Dbxref -> dbxref, Ontology_term -> ontology_term, as in gff3-contig/3.0.0 and later (from 4.0.0, issue 43); the generic pairs stay, and context CDS copies must agree too
round_trip:
  exact: byte-identical source recovery after projection and independent-original checks
  reconstruct: model fields plus semantic protein identity/attribute grouping mappings
  context: validation and export require the independent original context; supporting CDS/protein evidence is not exported as extra annotation rows
limitations:
  - requires explicit profile selection; never inferred from nine columns or identifier spelling
  - requires ID, HMMER source, PF accession, strand '.', phase '.', no source Parent tag, and one positive integer each in model_start and model_end
  - protein context must cover exactly the source references and contain one translation per CDS
  - no conversion of protein offsets into genomic positions is claimed
  - context provenance is declared; consistency checking does not authenticate external biology
  - edited-instance export and other NMDC annotation families are unsupported
