Curasynth / Data
The evidence graph, table by table.
Seven entities on stable identifiers, joined by 128 bridge tables and 237 declared relationships, from 43 sources whose license terms travel on every row. Every count on this page is measured from what is served.
Entities
One record per thing, one identifier that holds.
| Entity · identifier · spine | Records |
|---|---|
| Publications PMCID · PubMed Central open access, every paper in full text | 7,253,502 |
| Clinical trials NCT number · ClinicalTrials.gov, current to September 2026 | 601,387 |
| Biomarkers fid_biomarker · GWAS Catalog, CIViC, MarkerDB; human records only | 94,289 |
| Targets Ensembl gene · Open Targets, TTD, UniProt; structure from AlphaFold and the PDB | 80,177 |
| Diseases MONDO · human, current terms; ICD, SNOMED CT, MeSH, UMLS, OMIM, Orphanet, EFO, MedDRA and DSM-5 codes resolve to it | 24,030 |
| Drugs fid_drug, encoded from the ingredient set · Drugs@FDA, ChEMBL, UNII, RxNorm | 11,758 |
| Pathways fid_pathway · Reactome 2,870, WikiPathways 986, KEGG 372 | 4,228 |
Beside the seven: every reference in every held paper resolved to a work — 48.7 million in all, of which 41.4 million lie outside the corpus and 14.5 million of those are cited by three or more held papers — and 10.8 million authors and 580 thousand organizations, resolved through ORCID, OpenAlex and ROR and served as pages of their own.
Identity
Two keys are minted; everything else resolves to them.
A drug is its ingredient set
Every ingredient is standardized — an all-salt input is refused rather than registered under its counter-ion — and ranked in a fixed catalog; a drug's identifier encodes the set. A parent, its salt and a combination are distinct records that point at each other. When the catalog's ranking moved this September and nearly every identifier was re-issued, each bridge was re-keyed from its own ChEMBL id or InChIKey with the old key kept beside the new, and a release check now fails any table whose rows disagree with their own identity.
A disease is its MONDO term
MONDO is the hub almost every other entity resolves to. Nine code systems and free-text names map onto it through served crosswalks; a term MONDO retires redirects to its replacement, and a question about a disease reaches its subtypes through some 350,000 hierarchy relationships. Veterinary and deprecated terms are removed when the catalog is built, not filtered at query time.
Joined by
The bridge tables.
New evidence is added as a table beside the entity it describes, keyed on that entity's identifier. An index is never rebuilt to add a join, and the provenance of a row is the row.
| Literature and people | Rows |
|---|---|
| Paper-to-paper citations | 375,185,880 |
| References resolved to works | 301,416,662 |
| Citation intent and stance markers | 208,113,080 |
| Author–publication | 46,343,979 |
| Author–organization | 24,932,084 |
| Publication keywords | 26,990,338 |
| Publication funding | 6,356,461 |
| Publication–disease | 3,468,589 |
| Author metrics (h-index, citations by year) | 9,256,585 |
| Works resolved, held or not · with an OpenAlex record | 48,666,326 · 31,360,209 |
| Works outside the corpus cited by three or more held papers | 14,545,306 |
| Targets, drugs, trials | Rows |
|---|---|
| Target–disease, by evidence type | 3,431,585 |
| Target–disease, overall score | 3,148,612 |
| Target–tissue expression | 2,550,100 |
| Target–pathway membership | 1,067,629 |
| Adverse-event 2×2 tables · scored signals | 1,913,561 · 864,709 |
| Trial–disease · trial–drug | 819,136 · 290,483 |
| Trial parties, investigators, organizations | 5,072,886 |
| Binding measurements · consensus pairs | 461,919 · 145,727 |
| Gene Ontology · protein interactions | 336,582 · 236,837 |
| Disease hierarchy · phenotypes | 348,198 · 284,871 |
| Receptor occupancy · liability screens | 261,772 · 30,835 |
| Synthetic lethality · co-deletion · co-dependency | 250,480 · 874,159 · 447,650 |
| CRISPR dependency per cell-line model | 20,924,741 |
| Copy number per cell-line model | 4,008,458 |
| Dependency within a disease · within a lineage | 517,371 · 427,725 |
| Tumour expression by disease | 584,357 |
| Drug indications · ADMET endpoints · mechanisms | 59,234 · 50,839 · 7,179 |
| Pathway–disease | 183,423 |
| Structural domains · co-crystal evidence | 127,476 · 17,353 |
What one question can reach
Six kinds of evidence, and where each comes from.
LiteratureWhat has been written, and what it rests on
Every open-access paper in full text, and the citation graph between them — with the intent and stance of each citing sentence, so a claim can be followed back to what it cites and forward to who has confirmed or disputed it. Reference resolution joins the NIH Open Citation Collection and OpenAlex, which also reach the 14.5 million papers outside the corpus that three or more held papers cite; iCite supplies the relative citation ratio. Authors and organizations are pages of their own, with their metrics.
PMC · OpenAlex · NIH OCC · iCite · RORBinding and structureHow strongly, and where
Measured affinities from ChEMBL, BindingDB and the PDSP Ki database, collapsed to one consensus potency per drug–target pair. Co-crystal evidence from the PDB where a ligand has been solved with its target, domains and enzyme class from SIFTS, and a predicted or experimental 3-D structure for nearly every target, stored and rendered in place.
ChEMBL · BindingDB · PDSP · RCSB PDB · SIFTS · AlphaFold · PubChemGenetics and dependencyWhether a cell needs it
Open Targets association scores by evidence type; CRISPR dependency from DepMap at every level — the full 20.9M-row per-model matrix, and rolled up within a disease and within a tissue lineage with the contrast against every model outside it, so a gene essential in nine of ten of this disease's lines is not scored like one essential in a third of everything. Tumour expression per disease from TCGA, because a target the disease's tumours do not transcribe is not a target for it. Co-dependency, co-deletion and collateral-lethality candidates; published synthetic-lethal pairs; complex membership, paralogs and interaction partners.
Open Targets · DepMap · TCGA · SLKB · Complex Portal · STRING · EnsemblSafetyWhat it does to people, and what it hits
Twelve quarters of FDA adverse-event reports scored per drug and term; curated box warnings and withdrawals with their toxicity class; each drug's receptor-occupancy profile across an exposure ladder, screened against the standard in-vitro safety-pharmacology panel.
FAERS · ChEMBL warnings · MedDRA (gated) · Bowes et al., 2012ClinicalWhat has been tried
Every registered study linked to the diseases it enrolls and the drugs it tests, with arm-level effects recovered from posted results and pooled across trials of the same comparison; clinician-curated indications; sponsors, investigators and sites resolved to the same organizations that appear on the papers.
ClinicalTrials.gov · MED-RT · Drugs@FDA · RORExpression and pathwaysWhere it acts
Baseline expression for every target across 122 tissues at RNA and protein level, Gene Ontology annotation, and pathway membership from Reactome, WikiPathways and KEGG — each pathway linked to the diseases it is implicated in, with curated links marked apart from inferred ones and KEGG's research-only rows stamped as such.
Open Targets expression · GO · Reactome · WikiPathways · KEGG (gated)Sources and terms
Forty-three sources, and the terms travel with the row.
Every source is registered with its license, and every row it contributes is stamped with a distribution class: open, attribution, share-alike, research-only, license required, or per record where the terms are the article's own. A result assembled from mixed sources reports the most restrictive terms it inherits, and rows that a deployment is not licensed for are gated rather than served.
This is what lets one platform serve a public research deployment and a commercial one from the same tables: the row knows.
Open and attribution sources make up the spine: Open Targets, ClinicalTrials.gov, Drugs@FDA, FAERS, PubChem, UNII, RxNorm, MED-RT, PubMed Central, OpenAlex, ROR, iCite, MONDO, HPO, Ensembl, AlphaFold, the PDB and SIFTS, BindingDB, DepMap, TCGA, STRING, Gene Ontology, Reactome, WikiPathways, CIViC, GWAS Catalog, the Complex Portal. Share-alike: ChEMBL and SLKB. Research-only, served to research deployments: TTD, MarkerDB, KEGG, UMLS and ICD. License required, gated: SNOMED CT and MedDRA terms.
Checked, not assumed
Ten checks gate every publish.
Links resolve
Every declared relationship is walked: a cross-reference that no longer resolves fails the release rather than returning an empty page.
Counts are fresh
Each table's statistics sidecar is regenerated from the table itself; a missing or stale sidecar is a failure, so a quoted count is never older than its data.
Identity agrees
Drug-keyed rows are re-derived from their own ChEMBL id or InChIKey and must match the key they carry — verified by a second identity, never by mere existence.
Terms are stamped
Every row carries its source and distribution class; a part written without them does not pass the per-record license gate.
And a recovery set: published findings — binding affinities, deletion frequencies, dependency patterns, a collateral-lethality pair — beside what the platform measures for the same quantity from public data. Nineteen of twenty-three are recovered; the rest are carried in the table and marked, not dropped.
A look inside
Every name it has
Neighborhood
Trials landscape
Safety signals
Continue