EDCtox

Mechanistic evidence atlas. Scores summarise biological direction and magnitude. They are not a toxicity rank or a human-risk number.

Scoring methodology

The unit of record is a compound, in an exposure context, at a mechanism or outcome, supported by a finding. Domain columns are summaries of those records. They are not a substitute for them, and they are not added into one toxicity number.

  1. A signed score is biological direction and magnitude, from −4 (strongly protective or restorative) to +4 (strong disruption). It is not a human-risk number.
  2. Confidence (very low through very high) is stored separately. Human relevance (H0–H4) is stored separately. Neither is multiplied into the effect score.
  3. Null means no quantitative score was justified. Physiological high activity, such as melatonin at MT1/MT2, is not recorded as +4 disruption.
  4. Within an aggregation group, child mechanisms are one cluster. The cluster contributes its strongest absolute score, not the sum of its members.
  5. Across clusters in a domain, the headline is the strongest cluster, not the sum of clusters.
  6. When absolute scores tie, the cluster with higher human relevance is selected, then higher confidence. This is the only way human evidence is weighted more heavily than cellular evidence.
  7. Opposing directions are not averaged to zero. The headline keeps the larger magnitude, confidence drops one step when an opposing cluster is also present, and both sides stay in the trace.
  8. Research-only mechanisms, including reorganization energy, never enter the headline unless a later curated row changes that policy. ATP or oxygen changes are not treated as Marcus-theory evidence.
  9. The exposure-modifier domain does not contribute a profile column.
  10. Cross-domain pathway tags explain a possible causal chain. They do not collapse those domains into one total.
  11. Regulatory classifications are a separate layer and are not the scientific score.
  12. Seed hypotheses are labelled as such. They are starting assessments, not curated consensus.
  13. Activity magnitude is how strongly a pathway moves. The signed score is the protective or disruptive consequence, and it stays null when that consequence is not justified.
  14. Study quality belongs to an evidence finding. Evidence confidence belongs to the synthesized assessment. They are not the same field.
  15. Biological context (organ, tissue, cell, compartment) is stored separately so the same pathway can differ by place. Those rows are still capped inside one aggregation group.
  16. A dose metric names what a number would mean: prescribed dose, administered dose, dietary concentration, or an environmental concentration. Those metrics are not converted into each other.

What a cell is showing

−4 is strongly protective or restorative. +4 is strong disruption. Zero would mean no meaningful effect was demonstrated. A blank cell means no quantitative score is stored. “Phys.” means high physiological activity, such as melatonin at its own receptors, and is not a disruption score. Confidence and human relevance sit beside the number because potency is not human risk.

Biological activity and consequence

A compound can strongly affect a pathway without that activity being harmful. Melatonin at MT1 and MT2 is high physiological activity with no signed effect score. Activity magnitude records the movement. The signed score records a protective or disruptive consequence, and it stays empty when that consequence is not justified.

Study quality and confidence

Study quality, exposure assessment, confounding, and outcome measurement are properties of a finding. Evidence confidence is the judgment on the assessment those findings support. A finding can remain not assessed while the seed assessment still carries a confidence. They are not copied onto each other.

Biological context and dose

The same mechanism can differ by organ, tissue, cell type, and subcellular compartment. Those contexts are shared lookups. A prescribed dose, an administered experimental dose, a dietary concentration, and an environmental concentration are different metrics. They are not converted into one number. Human relevance stays independent of the effect score.

Gut and microbiome evidence

A shift in alpha diversity, beta diversity, taxonomic composition, or a Firmicutes:Bacteroidetes ratio is a description of the community. It is not a signed toxicity score. The ratio is a measurement, not a mechanism. Generic dysbiosis stays descriptive unless an assessment explicitly supports a functional or adverse consequence. Barrier permeability, LPS translocation, short-chain fatty acids, bile-acid signalling, FXR and TGR5, enteroendocrine signalling, and microbial bioactivation can carry a signed score when the evidence supports that function. Sequencing method, sample site, diet, antibiotics, probiotics, batch effects, and population change what a microbiome result means. They are method context, not a quality grade. Host–microbe xenobiotic mechanisms sit on the exposure domain because they change effective exposure or bioavailability. They do not add a profile column.

Endothelial and vascular evidence

ICAM-1, VCAM-1, eNOS, and endothelial reactive oxygen species are mechanisms. They are not a diagnosis of cardiovascular disease. Nitric oxide bioavailability, vascular tone, the endothelial barrier, angiogenesis, haemostasis, and microvascular function live on a vascular domain that is stored and is not one of the seven profile columns. Adhesion and vascular inflammation remain immune mechanisms, each in its own aggregation group. Endothelial reactive oxygen species remains a redox mechanism in its own group. An outcome such as endothelial dysfunction or a thrombotic event stays separate from the mechanism that might lead to it.

Aggregation groups and pathways

Mechanisms that share an aggregation group are one cluster. The domain headline uses the strongest cluster. It does not add the members. Two empty pathway definitions name a gut-barrier chain and a bile-acid / FXR / TGR5 chain. No compound is attached to either. Pathway membership records the relationship. Aggregation groups, not pathway membership alone, are what currently stop those steps from inflating a score.

External evidence

EDCtox is a synthesis layer. CompTox, ToxCast, ToxRefDB, ToxValDB, EDSP models, EASIS, ECHA, OpenFoodTox, CEBS, eChemPortal, IUCLID, AOP-Wiki, AICIS, APVMA PubCRIS, and PubChem stay external. An imported record does not change a domain score. A ToxCast hit is assay bioactivity. An AICIS inventory listing is industrial availability, not safety. An AOP mapping is a crosswalk, not evidence that every linked chemical causes the adverse outcome. ToxPi is a useful precedent for a multidomain picture. EDCtox does not collapse that picture into one weighted total.

  1. Exact structure or InChIKey, where a single structure is meaningful.
  2. DTXSID or another DSSTox curated identity.
  3. PubChem CID, checked against structure.
  4. EC number or ECHA substance identity.
  5. CAS Registry Number, as a useful and non-exclusive identifier.
  6. Name matching only as a manual review.

Language

In vitro results stay in vitro. Animal results stay animal. Observational human results are associations. A receptor assay is not a disease. Therapeutic dose is not an overdose. A parent compound is not its formulation. Intended pharmacology is not automatically an adverse effect. Contradictory findings stay visible. Regulatory text, when it is added, is a separate layer.