Human, mouse and dog protein abundance

The Oziolor et al. atlas provides parallel TMT protein measurements. ORFdata keeps each species in a separate dataset, using the same samples, assays, measurements and query API.

Dataset	Reported protein groups	Source channels	Source rows
pfizer-atlas-human-protein	10,352	37	325,702
pfizer-atlas-mouse-protein	12,124	63	615,991
pfizer-atlas-dog-protein	9,929	38	313,263

Release 2026-10-12.1 contains the complete March 23, 2026 summary tables from MassIVE MSV000099729, MSV000099754 and MSV000099744, respectively. Each source row contributes two observations: native iBAQ and rIBAQ. Neither is calibrated copies per cell. These are pooled tissue samples; rodent condition labels also include sex. Channels and technical runs do not count as independent donors.

pip install 'orfdata[data]'
orfdata datasets list --species dog
orfdata datasets download pfizer-atlas-dog-protein
orfdata datasets query pfizer-atlas-dog-protein --quantity protein_abundance --unit iBAQ --scale linear

from orfdata import DatasetRegistry, ScopedTarget

registry = DatasetRegistry()
dataset = registry.open("pfizer-atlas-dog-protein")  # After explicit download.
reference, = dataset.manifest.references
group = ScopedTarget(reference, "protein_group", "A0A0A0MPB0", "MSV000099744:ProteinGroup")
for value in dataset.measurements(target=group, unit="iBAQ", scale="linear"):
    print(value.sample_id, value.unit, value.value, value.status)

Omit the unit selector to return both native series. They retain the same source row and sample identity; two units do not imply two independent measurements.

Every target retains its species and the checksum of the deposited UniProt search FASTA. The original genomic assembly and UniProt release are unknown. They stay unknown; a future annotation mapping must be explicit and validated. Group representatives can be inspected through reference_mappings(), but their lookup does not establish all group members. Opting into unresolved mapped views keeps the original group measurement visible, without assigning its value to an individual protein. No RNA pairings are inferred in this release.

The complete search references are available through registry.download_reference(name) and registry.open_reference(name) without rebuilding abundance. This includes proteins without reported representatives; search membership is separate from detection or protein-group assignment.

Source NA values remain missing and numeric zeros remain measured zeros. Absent group/channel combinations have no observation. For dog, 64,039 combinations are absent and eight reported rows contain NA, reproducing the existing 64,047 missing iBAQ matrix cells. Each value retains its source line, column, original numeric text, representative, normalization sum and native Abundance field. Nonfinite transformed strings such as -Inf are kept as provenance only.

These searches used UniProt sequences without spliced isoforms. Coverage does not establish that an unreported ncORF was searched or absent, nor does bulk tissue MS establish MHC presentation. The article and deposit disagree on the instrument name; that discrepancy remains explicit in assay metadata.

Downloads are 87.3 MB (human), 154.8 MB (mouse) and 84.4 MB (dog). Installed SQLite files occupy 1.46 GB, 2.72 GB and 1.40 GB, respectively. Queries stream observations; they do not load the entire file or a dense matrix into memory.

Source tables are CC0 according to the MassIVE deposits; UniProt search records retain UniProt Consortium attribution and CC-BY-4.0 terms. Both sources have pinned hashes and sizes. The data artifact is separate from the Python package. Discovery and installed queries work offline. download verifies and installs the compressed artifact; it does not fetch the large vendor search database.

To reproduce a source import, use registry.build(name, source_paths=...), orfdata datasets build NAME, or orfdata.sources.pfizer.build_protein("dog"). The source overrides are protein.tsv and search.fasta; DataCache verifies their original checksums before running the importer. Source slices with actual positive, zero and missing values test the same importer for all three species.