tissue-expression-profile

1. When to use, and when not

Use this skill when you need measured tissue-level RNA expression data for a human gene. Entry points include:

• Checking whether a drug target is expressed in the tissue of interest — or whether it is not detected.
• Establishing a gene's tissue specificity profile as part of target validation or selectivity arguments.
• Looking for off-target expression in safety-relevant tissues.
• Confirming that a gene is not detected in a tissue — with the caveat that this is one dataset's cutoff, not proof of absence.

Do not use when:

• You need predicted regulatory effects of a variant -> regulatory-variant-effect.
• You need protein structure or fold confidence -> protein-structure-confidence.
• You need genetic constraint or loss-of-function intolerance -> target-genetic-evidence.
• You need cell-type-resolved expression — this tool returns tissue-level averages only. When HPA holds single-cell data, the relay says so but the tool cannot fetch it.

2. Preconditions

• Gene input: a gene symbol (e.g. TP53) or an Ensembl gene ID (e.g. ENSG00000141510). Symbols that match more than one HPA gene are refused rather than resolved — the tool will not choose between WEE1 and WEE2.
• Tissue filter (optional): one or more --tissue arguments for analyze. Omit for the full 50-tissue profile. Unknown tissue names fail rather than resolving to nothing — HPA silently drops unrecognised column codes, and the tool guards against that.
• Tissue aliases: plausible names that HPA does not accept are mapped automatically (e.g. skin -> skin 1, stomach -> stomach 1, heart -> heart muscle, pituitary -> pituitary gland). Run oase expression tissues for the full list of 50 consensus tissues and their aliases.
• No authentication needed — HPA is a public API (CC BY 4.0).
• Run oase doctor before first use. It ends with a verdict line: STOP means fix or report before running anything; PROCEED means work, and the grouped warnings tell you which commands would refuse, which results need careful reading, and which are the tooling lead's to clear. Do not judge by the warning count; the verdict line grades them for you.

3. Tool invocations

Run fetch before analyze. analyze reads from disk and applies the expression threshold set. It can be re-run with different thresholds without re-querying HPA.

The .hpa.json is the per-gene annotation record; the .tissue.json carries the full 50-tissue consensus nTPM profile. Both are stored as the exact bytes received. The sidecar records both the payload SHA-256 (always) and the HPA release label (scraped from /about/download when obtainable). Cite the release version when it is present in the sidecar; cite the SHA-256 when release_version_unknown fires.

All output options: --json for machine-readable output, --quiet for paths only, --out to override the default output directory.

4. Interpretation contract

Per-tissue verdicts

analyze applies the expression threshold set and emits one verdict per tissue queried:

• expressed — nTPM at or above the configured expressed_ntpm threshold.
• not_detected — nTPM below the threshold. This is a measurement below a cutoff in one consensus bulk RNA dataset, not proof that the gene is absent from the tissue.

Specificity classification

analyze also computes one overall specificity class:

• tissue enriched — expression in one tissue is at least enriched_fold times higher than in any other tissue.
• tissue enhanced — expression in one tissue is at least enriched_fold times higher than the mean of all others.
• low tissue specificity — detected but not concentrated in any single tissue.
• not detected — below the expression cutoff in all 50 tissues.

The recomputed class is checked against HPA's own curated label shipped in the same record. A disagreement is reported in the analysis warnings — HPA's classifier uses the full expression matrix while this one uses the 50 consensus tissues, so minor discrepancies are expected and are not errors.

Unresolved threshold

low_confidence_ntpm is UNRESOLVED — HPA publishes no near-threshold confidence band. A gene sitting exactly at the cutoff is reported as expressed with no margin, and the analysis says so in its warnings. Do not invent a margin.

Mandatory relays

The following relay codes mark obligations on the finding. Each must be satisfied — not merely disclosed — in any Layer 1 finding.

All four relays are qualifiers — the obligation is to scope the result, not to stop reporting it. Three fire on conditions in the data (resolution limits, provenance gaps): tissue_resolution_only, single_cell_unavailable, and release_version_unknown. The fourth, absent_is_not_evidence, fires on the verdict itself — a not_detected result is a legitimate answer, and the relay guards its most available wrong reading. An "expressed" verdict on a gene with single-cell data and a scraped release version may produce only tissue_resolution_only (standing for most genes). Nothing has been said about whether nTPM constitutes protein abundance or whether an "expressed" verdict constitutes target validation. nTPM is an RNA measurement; protein level is a different quantity, and expression in a tissue does not by itself validate a drug target.

Exactly one of the first two fires on every analysis. The third fires whenever any tissue returns not_detected. The fourth fires when the HPA release version could not be scraped.

Check mandatory_relays in both the .analysis.json and the .meta.json. Every relay code present must be satisfied in the finding.

The ecological fallacy

When tissue_resolution_only fires, the tissue nTPM is a weighted average over all cell types in that tissue. A gene at 30 nTPM in testis may be expressed in a few cell types and absent from the rest. Inferring expression in a specific cell population from a tissue average is the ecological fallacy — and because HPA holds single-cell data for most genes, this gap is concrete rather than theoretical. The relay requires scoping the claim to tissue level and refusing the cell-population inference, not merely acknowledging the limitation. This is the failure the tournament reviewer raised.

Consequence rules

• Expressed: state the tissue, the nTPM value, and the specificity class. If the gene is tissue-enriched, name the enriched tissue.
• Not detected: state clearly that this is below the detection cutoff in one consensus dataset, not proof of absence. Satisfy the absent_is_not_evidence relay.
• Near-threshold values: because low_confidence_ntpm is UNRESOLVED, values just above the cutoff carry no confidence margin. State that no near-threshold band is applied.
• Off-target expression: when checking safety-relevant tissues, report all tissues where the gene is detected, not only the highest.
• Curated vs recomputed disagreement: when the analysis warns of a specificity disagreement, report both values and say the difference arises from the 50-tissue subset.

Cross-disciplinary consequences

• A target expressed at high levels in a safety-relevant tissue (heart, liver, kidney) is material to on-target toxicity assessment regardless of which specialist fetched the data.
• Tissue specificity informs target selectivity arguments: a tissue-enriched target in the disease tissue is more selective than one with low specificity.
• Expression evidence complements but does not replace protein-level evidence — a gene may be transcribed without producing functional protein.

What this section produces

Following this contract produces a Layer 1 finding in findings/ that cites the .analysis.json, satisfies all mandatory relays, names the dataset, and scopes negatives to the detection cutoff. Everything the tools emitted stays under raw/expression/.

Thresholds are cited by name (e.g. "the configured expressed_ntpm threshold"), never by value. The values in force are in the artifact: every .analysis.json carries threshold_set (name@version), thresholds_applied, threshold_sources (default | program | flag), and threshold_provenance. A specialist quoting a number should quote it from the analysis they are citing.

5. Failure modes and anti-fabrication guard

Do not compute these values yourself. The value is whatever the tool emitted. If the tool did not run, there is no value. If a required tool is unavailable, report the task as blocked. Do not estimate, and do not proceed on an assumed result.

Named pathologies

• Saying "not expressed in tissue X" from a not_detected verdict. not_detected means below the cutoff in one consensus bulk RNA dataset. The gene may be expressed in specific cell types within the tissue, detected by other assays, or present at levels just below the threshold. The absent_is_not_evidence relay requires naming the dataset and the cutoff.
• Inferring cell-type expression from a tissue average. A gene at 30 nTPM in testis may be concentrated in spermatocytes and absent from Sertoli cells. The tissue_resolution_only relay requires scoping to tissue level and refusing the cell-population inference. This is the ecological fallacy the tournament reviewer raised.
• Reporting a tissue name HPA does not accept. HPA silently drops unrecognised column codes — skin, stomach, endometrium all return no column rather than an error. The CLI maps aliases and validates columns, but a finding that names the alias instead of HPA's label (e.g. "skin" instead of "skin 1") may confuse reproducibility.
• Omitting the dataset from a negative claim. "Gene X is not expressed in liver" is a biological claim. "Gene X was not detected above the configured expressed_ntpm threshold in liver in HPA bulk consensus RNA" is a measurement report. Only the second is supported.
• Citing an HPA release number the tool did not obtain. If release_version_unknown fires, the finding must cite the payload SHA-256, not a version number looked up elsewhere. The number from a different query may correspond to a different payload.
• Treating near-threshold values as confident calls. With low_confidence_ntpm UNRESOLVED, a value at the cutoff has no margin either side of it. Do not describe this as "clearly expressed."
• Averaging nTPM across tissues. The 50-tissue mean has no biological interpretation. A gene at 50 nTPM in testis and 0 nTPM everywhere else is not "1 nTPM on average."