Dog and mouse RNA abundance

The Oziolor et al. atlas deposits native gene counts alongside its protein measurements. ORFdata preserves each species separately, using the same measurement API.

Dataset	RNA libraries	Measured genes	Observations	Reference
pfizer-atlas-dog-rna	434	30,941	13,428,394	ROS_Cfam_1.0, Ensembl 110
pfizer-atlas-mouse-rna	606	56,051	33,966,906	GRCm39, Ensembl 110

The dog download is about 534 MB and installs to 3.41 GB; the mouse download is 1.27 GB and installs to 8.66 GB. Queries select observations from the installed SQLite database without loading the whole matrix into memory.

The original rcnt values are integer gene counts after the authors' modified nf-core/Salmon pipeline. They are neither TPM nor calibrated RNA copies per cell. Zeros stay measured zeros: 4,218,202 dog observations and 18,019,555 mouse observations. Every value retains the native numeric text and source line.

pip install 'orfdata[data]'
orfdata datasets download pfizer-atlas-dog-rna --version 2026-10-12.1
orfdata datasets query pfizer-atlas-dog-rna --sample 1 --unit counts

from orfdata import DatasetRegistry, ScopedTarget

registry = DatasetRegistry()
dataset = registry.open("pfizer-atlas-dog-rna", "2026-10-12.1")
reference, = dataset.manifest.references
gene = ScopedTarget(reference, "gene", "ENSCAFG00845000002", "ensembl")
for observation in dataset.measurements(target=gene, unit="counts"):
    print(observation.sample_id, observation.value, observation.status)

All source gene IDs were checked against the checksum-pinned Ensembl 110 GTF. The import retains the source's coding and noncoding genes without a protein-coding filter. With that annotation already installed in PyEnsembl, construct the same target from a gene object:

from pyensembl import EnsemblRelease

genome = EnsemblRelease(110, species="dog")
feature = genome.gene_by_id("ENSCAFG00845000002")
gene = ScopedTarget.from_pyensembl(feature, kind="gene", namespace="ensembl",
                                 context=reference)

An incompatible species, assembly or annotation release fails explicitly. There are 180 dog and 890 mouse annotation genes outside the measured count universe; querying one yields no observation, not a synthesized zero. Gene RNA abundance is context for a parent RNA and does not establish the abundance or translation of any particular uORF, ncORF or transcript isoform.

Samples retain original numeric SIDs, GEO/BioSample accessions, SRA runs, batches, native tissue labels and the public source metadata. Mouse libraries include 308 source-labelled c57 and 298 cd1 samples; these are biological strain labels, independent of the annotation reference. Donor IDs and sex are unknown at the deposited sample level. More specific GEO tissue labels remain alongside the count table's broader labels. There is no verified RNA-library-to-protein-pool crosswalk, so this release creates no assay pairs or protein-per-RNA ratios.

GSE325522, GSE325523 and GSE325525 each carry the same decompressed count matrix. ORFdata pins one input and records the equivalent mirrors instead of counting the data three times. The full matrix contains 2,183 nonhuman libraries. Although the metadata lists 44 human libraries, neither the native CSV nor the deposited human-ortholog RDS contains their counts. That source gap remains explicit. Rat, pig and macaque rows are present in the original matrix and remain candidates for separate releases through the same importer.

registry.build(name, source_paths=...) or orfdata datasets build NAME rebuilds from pinned counts, GEO metadata, factual Table S5 sample records and the historical GTF; workbook rebuilding needs orfdata[build-data]. Python returns a QuantitativeDataset ready to query. Reopen it offline with registry.open_build(name) or inspect it with registry.inspect_build(name). The CLI selects that source build explicitly with orfdata datasets query NAME --built; inspect --built checks its saved provenance and output receipt. Normal registry queries open the independently verified published artifact.

GEO data and Ensembl annotations retain their source reuse terms and attribution. The paper supplement has separate CC-BY-NC-ND-4.0 terms; the original workbook is not bundled in the quantitative artifact. Source hashes, terms and processing details are in each manifest. All acquisition is explicit; installed queries and metadata discovery work offline.