Target specificity
Do cancer targets differ in how specific they are to cancer? Yes: some are alterations only tumour cells carry, some are ordinary proteins the tumour has more of, some are shared with one normal cell lineage, some are everywhere, some are inherited, and some sit on immune cells rather than the tumour. 904 targets with an approved or clinical-stage medicine carry a class, read from label readouts, drug mechanisms, the Human Protein Atlas and UniProt; each target page says which data decided it.
Tumour-specific alteration
235 targets →The medicine aims at an altered form of the protein (a mutation, a fusion or a neoantigen) that normal cells do not carry.
For a medicine: Normal cells lack the altered form, so on-target harm to healthy tissue is low; a tumour or blood test must find the alteration first.
Tumour-associated overexpression
71 targets →Normal tissue carries the target too, at lower levels; the medicine relies on the tumour carrying much more of it.
For a medicine: Normal tissue with the target is hit too (on-target, off-tumour toxicity), so the dose and the level of the target in the tumour both matter; an expression test often decides eligibility.
Lineage antigen
31 targets →The target sits on one normal cell lineage (B cells, plasma cells, myeloid cells) as well as on the cancer that grew from it.
For a medicine: The medicine clears the normal lineage as well (B-cell aplasia, low antibodies, cytopenias); that is tolerated when the lineage can be replaced or its job supported.
Broadly expressed or essential
268 targets →Nearly every dividing cell carries or depends on the target; the medicine works because tumour cells divide faster or depend on it more.
For a medicine: Side effects follow the body's fastest-dividing normal cells (marrow, gut lining, hair); no target test is needed.
Germline variant
22 targets →The alteration is inherited and present in every cell of the body; it shapes cancer risk and picks medicines that exploit the loss.
For a medicine: A blood test finds it, it may run in the family, and the same result picks synthetic-lethal medicines such as PARP inhibitors.
Immune or microenvironment target
25 targets →The target is on immune, stromal or bone cells around the tumour, not on the tumour cell itself.
For a medicine: Harm comes from the immune system acting on normal organs (immune-related adverse events), not from the drug hitting tumour tissue.
How many cancer types
Counted over cancer families with a prevalence row, a label threshold or a catalogue link; Open Targets associations are the fallback.
Evidence for this target sits in one cancer family: the prevalence rows, label thresholds and approvals all point to one place. Testing for it belongs to that cancer's standard work-up.
The target matters in a handful of cancer families, each with its own evidence. Testing is routine in those cancers and worth asking about in related ones.
The target appears across many cancer families, and often a tissue-agnostic approval lets any tumour carrying it be treated. Broad tumour profiling can find it in any cancer; a tissue-agnostic label means the medicine may be used wherever it is found.
No prevalence row, threshold, approval or catalogue association ties the target to a cancer type yet. Ask whether a profiling panel reports it; no cancer-specific guidance exists yet.
Tumour-agnostic approvals
10 targets →A label lets a medicine aimed at these targets treat any solid tumour that carries the alteration or expression level, whatever organ it started in (NTRK fusions, RET fusions, BRAF V600E, HER2 IHC 3+, MSI-H or dMMR with the PD-1 antibodies). Broad tumour profiling is how they are found.
How the classes were assigned
scripts/fetch-target-specificity.ts ran on 2026-09-24 over every target with an approved or clinical-stage medicine, applying rules in a fixed order and writing which rule fired into each target's note. Immune checkpoints and stromal targets read immune or microenvironment from the record's class. A tumour suppressor or DNA repair gene whose UniProt disease text names a hereditary cancer syndrome reads germline variant (TP53 is excluded by name: its mutations are overwhelmingly somatic). The label readouts filed under a target decide next: a majority measuring a sequence variant reads tumour-specific; a majority scoring protein level or gene copies reads lineage antigen where the Human Protein Atlas finds the gene enriched in a blood lineage, else tumour-associated overexpression. Then the mechanisms of the medicines aimed at an oncogene driver, then the antigen-directed medicines (ADCs, CAR-T, engagers, radioligands, tracers) against a membrane antigen, then catalogue driver calls without a corpus medicine, and last the atlas's RNA tissue specificity: low specificity or an essential protein reads broadly expressed. Where none of that decides, the target reads "not established" and the note says what was seen; 252 targets sit there, nearly all catalogue genes with no corpus medicine and a tissue-enhanced expression pattern that says where the protein lives but not whether the tumour differs from normal tissue.
Distribution counts cancer families (the top of each cancer's parent chain, with blood cancers merged into leukaemia, lymphoma and myeloma) that carry a prevalence row, a current label threshold or a catalogue link for the target; one family reads one type, two to four a few, five or more many. A tissue-agnostic approval or threshold reads many types and sets the tumour-agnostic mark.
Sources: label readouts and thresholds on the biomarker pages; drug modalities and mechanisms on the product pages; the Human Protein Atlas version 25.1 downloadable data (RNA tissue, cancer and blood lineage specificity, normal tissue and cancer antibody staining), used under CC BY-SA 3.0; UniProt involvement-in-disease text (CC BY 4.0); Open Targets disease associations (CC0). No licensed list was used. Machine-readable: all targets as JSON (fields specificity, distribution, tumourAgnostic, specificityNote, specificitySources).