1Platform & data sources

Q-omics is a pre-computed, consensus-based cancer multi-omics platform — most results are validated across many sampling combinations before they're ever queried, not computed fresh per request (§4 explains why that matters).

Two more domains sit outside this table because they use their own cohort system rather than TCGA/CCLE/CPTAC lineage codes: Pembrolizumab (6 clinical-trial cohorts, J-codes — §7) and DIG (NCI60 cell lines, 13 lineages — §5/§7).

2Data types

Patient / tissue domain

Cell-line domain

3RPPA vs. mass-spec protein

Two genuinely different protein data types exist side by side — mixing them up is the single most common protein-data mistake to guard against in client code.

4The consensus mechanism

The single most distinctive thing about how Q-omics results differ from a raw statistical query.

Most results are pre-computed across many sampling combinations (stage, gender, sample split, time period — the exact factor set varies by analysis) before being stored. A result surfaces because it held up across many arbitrary re-splits of the data, not because it happened to be significant on one particular split.

High S_Cscore → not a statistical artifact of one arbitrary split. High L_Cscore → broad pan-cancer biology rather than a lineage-specific quirk. The two together let a consumer prioritize hits that are both reproducible and biologically general. Exception: Correlation and DIG rely on direct statistical metrics — no consensus sampling behind those two.

Two data-access tiers follow from this: pre-computed results (sampling-consensus validated, fast, what List/browse endpoints return) vs. on-the-fly computation (any valid entity × lineage pair the raw data supports, whether or not a significant pre-computed result exists — what Plot endpoints compute live). The only hard wall is whether the entity exists in the underlying data files at all.

5Analysis catalog

Nine analysis families. For each: what it answers, which domain, and the output-format rule — the single biggest thing to get right when building against this data, since List and Plot pull from genuinely different sources (§4).

1 — Survival

Whether a molecular feature associates with prognosis, measured as Kaplan-Meier AUC (1 = full survival, 0 = immediate non-survival) rather than a bare p-value. Tissue only. Up to 216 sampling combinations per gene. List shows pre-validated hits only — every row already cleared the significance bar.

2a — Normal vs. Tumor

Matched-pair differential expression, tissue only, 18 lineages with matched normal samples (BLCA, BRCA, CHOL, COAD, ESCA, HNSC, KICH, KIRC, KIRP, LIHC, LUAD, LUSC, PAAD, PRAD, READ, STAD, THCA, UCEC). Mutation and RPPA have no NT table — not available for this analysis at all.

2b — Drug response (Responder vs. non-responder)

RECIST-based responder classification against 30 patient drugs (§7 has the per-drug lineage table). Never confuse with Association — patient drug response is exclusively a Difference-analysis input; "mutations associated with Cisplatin resistance" routes through cell-domain Association using CCLE drug response instead, a different dataset entirely from the clinical RECIST one.

3a / 3b — Association & Correlation

Same infrastructure, different questions. Association tests both directions independently (X-score / Y-score, two p-values) and degrades gracefully to a hit table when an entity is left open. Correlation is always a scatter plot — both entities and a specific lineage are required, no open-axis table form exists. Both support tissue and cell domains with materially different valid-combination rules (§6).

4 — Pembrolizumab response

Same Responder/Non-responder structure as 2b, but its own clinical-trial cohort system (J0–J5, not TCGA/CCLE codes) and restricted to RNA expression / infiltrating cells as the paired data type. The only checkpoint inhibitor with cohort data in the platform (§10).

5a / 5b — Synthetic lethality

Gene pairs where joint disruption is more damaging to the cancer cell than either alone — the BRCA/PARP-inhibitor logic, generalized. Cell (5a): RNA × CRISPR/shRNA only. Patient (5b): D1/D2 must share the same type (RNA×RNA, MS×MS, or immune-cell×immune-cell) — but the List form only ever has RNA×RNA data (the pre-computed SL survival table stores RNA only); MS and immune-cell pairs only return something when both entities are pinned as a Plot.

6 — DIG (drug-induced gene expression)

Expression fold-change at 4 dose×time conditions (low/high × 6h/24h) for 15 specific drugs against NCI60 cell lines. Cell domain only. Only ~12–34% of genes have DIG data (2,800–8,500 of ~38,000, varying by drug) — worth flagging since it's easy to assume full gene coverage by analogy to RNA expression.

7 — Pan-cancer profile

One entity's distribution across every lineage at once, box-plot-per-cancer-type. Requires a specific entity — there's no pan-cancer profile without one. Not the same thing as an ALL-lineage List (which aggregates significant hits across lineages, a different question and a different output shape).

9 — Neoantigen analysis

More precisely: proteogenomic tumor-antigen discovery. Not strict neoantigen prediction (no mutation→MHC-binding→peptide pipeline) — instead, which mass-spec-quantified proteins are over-expressed in tumor-vs-normal or mutant-vs-wildtype, as candidate immunotherapy targets. Mass-spec protein abundance only, CPTAC tissue only, and lineage is ALL only — no per-lineage neoantigen queries in the current schema.

6Valid data-type combinations

Association and Correlation each have different valid-pair rules per domain — this is the table to check before assuming a combination exists.

Tissue Association (37 valid pairs)

Excluded: Mutation×Mutation, RPPA×MS (either direction), RPPA×GO(Protein) as D1, GO×GO, Mutation×GO (either direction).

Cell Association

Wider than tissue — adds Drug response, CRISPR, shRNA, and cell-line-only Methylation. Rule of thumb: everything is valid except GO×GO (4 combinations) and Mutation×GO (4, either direction). Self-pairs allowed for every continuous type; Mutation×Mutation is allowed only as a Table (both entities null — a binary×binary box-plot has no meaning).

Tissue Correlation (29 valid pairs)

No Mutation (binary, not continuous — Correlation needs two continuous values). Excluded: RPPA×MS (either direction), RPPA×GO(Protein) as D1, GO×GO.

Cell Correlation

RNA, MS, RPPA, Drug response, CRISPR, shRNA, cell-line Methylation, GO(R), GO(P) — 9 types, everything valid except the 4 GO×GO pairs. Mutation excluded (binary); infiltrating cells absent from the cell domain entirely.

7Lineage coverage

Coverage is not uniform across lineages — it depends on the (data type, analysis) pair. This is the exhaustive set, not a suggestion list; there is no data outside it.

Tissue — by data type × analysis

Survival with RNA/Mutation/IC/RPPA/GO(RNA) — any of the 34 TCGA lineages, or ALL. Survival with mass-spec/GO(Protein) — the same 9-lineage CPTAC pool as above.

Cell line — by domain

Per-drug lineage availability — patient drug response (30 drugs)

A lineage requested that isn't in a given drug's list has no data for that drug — the closest available lineage should be substituted rather than falling back to ALL, which understates specificity.

8Clinical-term mapping

Canonical stage/age bucketing (AJCC / WHO conventions) — useful reference for any client building a clinical-stratification UI on top of the API.

9Entity coverage at a glance

10Known limitations

What Q-omics doesn't do, stated plainly — coverage gaps worth checking before assuming broader support than exists.

1Data type tokens

The x_type (and data_type) parameter selects what kind of measurement you're querying. These are the literal string values the API accepts — not the display names used in Part I.

Legacy vocabulary aliases

Older display names are accepted and silently normalized to the tokens above — send either.

Sample domain

2Dataset resolution

The dataset parameter names the cohort. If omitted, it's auto-resolved from sample_type + x_type:

CPTAC-only types (force dataset=CPTAC even without explicit lineage): protein_ms, go_protein_ms, protein_ms_normal, go_protein_ms_normal, neo_wt, neo_mt, neont_wt, neont_mt.

Two datasets sit outside this auto-resolution and must be named or implied by lineage code instead: PEMB (Pembrolizumab cohorts, J-codes) and NCI60 (DIG, drug-induced expression).

3Lineage codes

Pan-lineage tokens

Any of the following (case-insensitive) means "every lineage in the domain" — used for ALL-lineage List queries and Pan-cancer Profile:

'', all, j0, pemb_all, tcga_all, nci60_all, none

4Filter parameters

Gender

Stage

Quantile split

Clinical-term mapping (for stratification UIs)

5Valid combinations

Association and Correlation each accept a different set of x_type/y_type pairs per domain — the two matrices below are load-bearing, not illustrative.

Tissue Association — 37 valid pairs

Tissue Correlation — 29 valid pairs (no mutation)

Cell domain — rule form

By-analysis valid x_type sets

6Entity resolution

Response classification, RECIST → ordinal (patient drug response, DIG-adjacent fields):