Skip to content

NCBI fetch accepts PDB structure-chain accessions

Motivation

In the New Search Generate query from NCBI modal, searching mpox (db=nuccore) returned PDB structure chains such as 8WGZ_T (the 5-bp ssDNA chain T embedded in the MPOX E5 hexamer structure). Selecting one and pressing Load genes / Insert sequence failed with:

accession is not a recognisable NCBI identifier

The fetch-path validator (normalise_accession) only accepted GenBank/RefSeq accession shapes ([A-Z]{1,4}_?[0-9]+...), so a digit-led PDB ID + chain (8WGZ_T) was rejected even though the same search legitimately surfaced it — an inconsistency between what the search returns and what the fetch accepts.

User-facing change

  • GET /api/ncbi/nuccore/{accession} (summary / genbank / fasta) and the modal's Load genes / Insert sequence now accept PDB structure-chain accessions of the form <4-char digit-led PDB ID>_<chain> (e.g. 8WGZ_T, 1ABC_AB), in addition to the existing GenBank/RefSeq shapes.
  • The conservative rejection of arbitrary strings is unchanged — only the PDB chain shape was added, still capped at 32 characters and still upper-cased.

API/IaC diff summary

  • api/services/ncbi/nuccore.py — added _PDB_ACCESSION_RE + an _is_recognised_accession() predicate used by both the pipe-extraction loop and the final check in normalise_accession.
  • No frontend change (the modal forwards the accession unchanged; the error originated in the backend validator).

Validation evidence

  • uv run pytest -q api/tests/test_ncbi_nuccore.py → 65 passed (new accepts: 8WGZ_T, 8wgz_t→8WGZ_T, 8WGZ_T, 1ABC_AB).
  • uv run ruff check api/services/ncbi/nuccore.py → clean.