Evidence methodology

Names are labels. Accessions and sequences are identity.

Protein identity is modeled as a UniProt accession, taxon, sequence version or checksum, and canonical-versus-isoform state. PDB structures remain identified by PDB entry, polymer entity, and chain, then map through SIFTS with observed coverage, mutations, deletions, tags, and chimeric boundaries preserved.

Family membership uses the curated inclusion list plus supporting alpha-carbonic-anhydrase domain evidence. A name substring or shared reaction is not sufficient.

Every assertion returns to a release and record.

Release provenance records each build input's snapshot path, canonical request URL, parameters, release/version, retrieval timestamp, response hash, upstream IDs, and license context. The corpus producer retains those source snapshot files separately; they are not embedded in the served binary. Source-reported measurement fields remain intact in the downloadable corpus, and normalized additions carry separate transform provenance rather than overwriting them.

Inspect the current source ledger.

Joins must survive exact context.

• ChEMBL target admission requires a direct SINGLE_PROTEIN target, matching taxon, and exact component accession.
• Compounds use full stereochemical identity when relevant. Parent and salt forms remain distinct measurements.
• Constructs keep mutations, truncations, fusions, tags, missing regions, and sequence coverage.
• Raw measurement value, relation, and unit are preserved alongside distinct source-standardized fields.
• Predicted structures remain a separate evidence class from experimental structures.

The decision engine is strict by design.

The server checks required identity and numeric-value fields before any matching. It then evaluates target scope, taxon, construct, catalytic status, endpoint type, relation operator, unit, assay format, pH, temperature, substrate, buffer, tested molecular form, stereochemistry, duplicate dependence, reaction direction, evidence class, assay or matched-protocol group, and provenance completeness.

No Ki/Kd/IC50/Km/kcat conversion is inferred. Censored bounds remain bounds. Same-paper records are not automatically comparable.

A reaction hypergraph with a designed projection.

A release-pinned, source-linked projection of selected human metabolic pathways spanning carbohydrate, energy, lipid, amino-acid, nucleotide, cofactor, porphyrin and biological-oxidation chemistry.

The map shows curated reaction direction and compartment identity, not measured flux, concentration, tissue dominance or thermodynamic magnitude. Animated traversal marks a selected route only.

The canonical model treats each Reactome event as a node with role-bearing input and output participants. Compound instances retain Reactome identity, ChEBI identity when supplied, formula, entity class, and compartment. Catalysts, Rhea cross-references, literature citations, pathway membership, and source URLs remain attached to the event rather than being inferred from line proximity.

• Overview shows 12 domain territories and 49 pathway neighborhoods.
• Pathway level exposes reaction nodes and primary non-currency compounds.
• Reaction level adds stoichiometric participants, cofactors, sets, complexes, source context, and citations.
• Transport is marked only when source participants occupy different compartments; no membrane crossing is inferred from layout.
• Collapsed objects retain their Reactome class and never receive an invented balanced equation.

The direction-aware route finder traverses admitted input-to-output relationships. ATP, NAD(H), NADP(H), water, protons, phosphate, and free CoA stay visible but are excluded as intermediate route shortcuts. Selected-route motion indicates order only and stops when reduced motion is requested.

Oxalosuccinate is shown as charge-specific CHEBI:153066 inside the NADP-dependent IDH2 mechanism, paired with the human net event R-HSA-450984 and Rhea reaction RHEA:19630. It is not drawn as a freely accumulating pool.

Corrections supersede; they do not disappear history.

Every manual identity decision, matched-protocol group, correction, and source update should record reviewer, date, rationale, and the superseded state. Material changes enter the public changelog. A future expert review will be labeled only after it occurs.

What this release does not claim.

• No kinetic or flux simulation.
• No tissue-specific concentration or direction claims from topology alone.
• No inference from co-crystal presence to binding affinity.
• No universal quality ranking across X-ray, cryo-EM, NMR, and predicted structures.
• No clinical efficacy, safety, diagnosis, dosing, treatment, or personal guidance.
• No independent expert approval until an external biochemist completes review.