Skip to content

Matched human tissue RNA and protein abundance

The Wang et al. atlas measures adjacent cryosections from 29 healthy human tissue specimens. wang-human-tissue-abundance keeps six original table series in the same species-scoped API as dog, mouse and yeast data.

Source table Targets Native quantity
EV1 protein groups 15,210 iBAQ
EV1 genes 13,640 Author-normalized iBAQ
EV2 transcripts 93,674 FPKM
EV2 genes 18,072 Author-normalized FPKM
EV5 protein groups 15,210 Estimated copies/cell
EV5 transcripts 60,519 Estimated copies/cell

The download is approximately 397 MB and installs to a 3.86 GB SQLite database.

pip install 'orfdata[data]'
orfdata datasets download wang-human-tissue-abundance --version 2026-10-12.1

from orfdata import DatasetRegistry, ScopedTarget

registry = DatasetRegistry()
dataset = registry.open("wang-human-tissue-abundance", "2026-10-12.1")
reference, = dataset.manifest.references
gene = ScopedTarget(reference, "gene", "ENSG00000004059", "ensembl")  # ARF5
for value in dataset.measurements(target=gene, sample_id="V122"):
    print(value.quantity, value.value, value.unit, value.status)

Select the same gene on both sides for a specimen's native gene-level comparison:

match, = dataset.matched_measurements(
    allowed_bases=["adjacent_donor_tissue"],
    pair_ids=["rna_genes:protein_genes:V122"],
    left_target=gene, right_target=gene,
    left_filters={"unit": "FPKM"}, right_filters={"unit": "iBAQ"},
)
print(match.left[0].value, match.right[0].value)  # 51.09, 112559620.9

Specimens and source cells

There are 6,273,425 source cells. The same specimen has several processing views; these are not additional biological replicates. Native specimen IDs, RNA/MS run IDs, units, source cells and processing methods remain attached to the data. Bone marrow has no matched RNA column and is outside the 29-tissue dataset. Every source cell and RNA-copy input link was independently reconciled before publication. Missing or censored cells remain explicit observations.

PyEnsembl targets and search references

With Ensembl 83 already installed in PyEnsembl, the same target can be built with ScopedTarget.from_pyensembl(gene_object, kind="gene", namespace="ensembl", context=reference). The context pins the historical GRCh38.p5 assembly accession, GTF and complete Ensembl 83 protein FASTA. Alternate-reference protein entries absent from the GTF remain in the native search set. Search lookups preserve versioned FASTA IDs; they do not silently strip versions or imply detection.

Interpreting the values

Protein-group values remain group observations. Representative proteins and their parent genes/transcripts are context, with unresolved individual attribution. The source's gene-level summaries permit gene-level RNA/protein comparisons; they do not quantify a particular isoform, uORF or ncORF. Pairings explicitly use adjacent_donor_tissue, independently of the target relationship. See the comparison API for selecting pairings and deriving ratios.

Protein copies are the authors' proteomic-ruler estimates, not isotope-standard measurements. Raw intensity inputs and per-sample calibration constants are absent from these tables; corresponding iBAQ observations are linked as associated data, not falsely recorded as the ruler's inputs. RNA copies link their source FPKM cells and retain the assumption that mRNA constitutes 2% of total RNA mass. Unreported calibration uncertainty stays unknown.

Protein zeros are unquantified values; gene-RNA and transcript-copy zeros retain the source censoring semantics. They are not measurements of biological absence. The methods state a 1-FPKM threshold, but gene values undergo median normalization; 2,677 positive gene values are below 1. Their values remain intact, and no numeric detection limit is invented in normalized units. Unfiltered transcript FPKM retains its native zeros and tiny numeric text values; a nonzero estimate does not establish expression above the study's 1-FPKM cutoff. Original numeric text and zero sentinels remain in provenance. Stable identifiers join the source tables even when their display gene labels differ.

The Eraslan reanalysis selects major isoforms and normalizes exon/intron counts differently. It is a separate planned view of the same specimens, not silently mixed into this release. These tables cover canonical coding-gene expression; they add RNA/protein context without expanding the experimental noncanonical-ORF catalogue.

Source supplements retain CC-BY-4.0 attribution, and Ensembl annotations retain their source terms. Discovery and installed queries work offline; acquisition and source rebuilding are explicit through DataCache.