Native protein search references

A search reference lists the sequences available to a particular MS search. It is independent of reported abundance or experimental evidence that a protein exists. ORFdata exposes the exact deposited FASTA, including unreported entries.

Dataset	Search entries	FASTA bytes, downloaded and installed
pfizer-atlas-human-protein	20,600	13,831,073
pfizer-atlas-mouse-protein	21,989	14,325,720
pfizer-atlas-dog-protein	20,649	14,659,832

These are uncompressed, no-isoform UniProt search sets from data release 2026-10-12.1. The historical UniProt release and genomic assembly remain unknown. ORFdata does not replace them with a current proteome.

from orfdata import DatasetRegistry, ScopedTarget

registry = DatasetRegistry()
name, version = "pfizer-atlas-dog-protein", "2026-10-12.1"
spec, = registry.search_references(name, version)  # Metadata only; no I/O.
registry.download_reference(name, version)        # Explicit verified download.
search = registry.open_reference(name, version)   # Offline; raises if missing.

entry = search.get("A0A068LEV2")
print(entry.header, entry.sequence, entry.sequence_sha256)
assert search.membership(entry.target).status == "present"
for entry in search.entries():
    assert entry.target.reference == search.reference

absent = ScopedTarget(search.reference, "protein", "not-in-this-search", "uniprot")
assert search.membership(absent).status == "not_in_search"

Dog has 9,929 reported group representatives and 10,720 other search entries, including A0A068LEV2. Those other entries have no reported representative; that does not prove that their peptides were undetected. The summary table does not resolve all protein-group members. Query group abundance separately through dataset.measurements().

Membership accepts an explicitly scoped native protein target. A different reference raises an error. Other namespaces, ORFs, genes and protein groups return unknown: first supply an explicit, validated reference mapping. Search membership never creates a zero, allocates a group's abundance, establishes protein detectability, or supplies MHC presentation evidence. Exact accession lookup does not perform sequence matching or infer coverage of a new ORF.

Each entry retains its full header, FASTA ID, residues, sequence hash, source taxon, optional gene label and source line. search.spec.to_dict() records the dataset version, reference identity, sequence-set SHA-256, source URL, size, licence and attribution. The deposited records do not specify decoy/control roles; those fields stay None. UniProt records retain CC-BY-4.0 attribution.

orfdata datasets references pfizer-atlas-dog-protein --version 2026-10-12.1
orfdata datasets download-reference pfizer-atlas-dog-protein --version 2026-10-12.1
orfdata datasets inspect-reference pfizer-atlas-dog-protein --version 2026-10-12.1
orfdata datasets query-reference pfizer-atlas-dog-protein --version 2026-10-12.1 --target A0A068LEV2

Pin the quantitative dataset version and retain the matching search spec when exporting an analysis. Both use the same native ReferenceContext; select reference=spec.reference and asset_id=spec.source.asset_id if a dataset has multiple declared search sets (--reference KEY --asset ID in the CLI). Reference downloads use a separate DataCache store and never rebuild or download abundance. Local source overrides are checksum-verified; force=True creates a new immutable bundle and preserves already-open readers. Ordinary discovery, inspection and opening stay offline. Published quantitative hashes are unchanged.

The same reference model accepts any taxon, with no inferred cross-species joins. Adapters must explicitly declare their native search sets and formats. The initial parser supports deposited UniProt FASTAs. CoPY yeast's QconCAT FASTA is synthetic calibration material, so it is not exposed as a native yeast proteome; that historical search reference remains an explicit source gap.

Historical Ensembl searches

Declared Ensembl peptide FASTAs retain exact versioned identifiers and native headers, including alternate-reference entries and stop characters. Gzip source hashes identify the original compressed bytes. An Ensembl header does not encode taxonomy: source_taxon_id is None, while the explicit pinned reference retains the source species. Gene labels use the native gene symbol when present, otherwise the native versioned gene ID. No ID-version stripping or protein-detection claim is implicit in a search lookup. See the human tissue dataset.