ExposomeAtlas — Application Guide

Access & sign-in

The public front page, sign-in and researcher registration.

Description: The front page (index.html) is public: it introduces the resource, shows live corpus counts, the direction-of-effect split and the top exposure categories, and offers eight entry points into the evidence. Exploring the full corpus needs an account.

What you see

• Sign in: the sign-in dialog opens from the front page; a signed-in user lands on the Dashboard. A session ends after a few minutes without activity and after 12 hours in any case.
• Request access (register.html): full name, institution (search the list or type a new one), work email and password. The researcher receives an email confirming the request and the administrators are told by email; one of them approves or rejects it on the Users screen, and an approved researcher is told by email and can sign in at once. (Emails need a mail server in the console's settings; until one is configured the Users screen says so and every notice that was not sent is listed under Sign-in activity → Email.)

Persistent identifiers need no sign-in: every exposure, outcome, association and paper has a link under /exposomeatlas/id/… that opens a landing page in a browser and returns linked data (JSON-LD) to software.

Overview

The two summary screens.

Dashboard (Evidence overview)

Description: The home screen — a single-scroll summary of the whole corpus: the curated literature associations together with the NHANES exposome-wide atlas layer.

Purpose: Get the corpus-wide picture at a glance — how much evidence exists, its direction and quality split, which exposures and outcomes dominate, and the newest records — before drilling into a specific view.

What you see

• Lead line and split pill: the share of records that are quantitative vs direction-only, and an “Open Association Explorer” button.
• Clickable KPI tiles: Papers · Associations · Exposures · Outcomes · Mechanisms · Cohorts · Countries.
• Evidence base composition: a stacked bar of quantitative vs direction-only records; a stat strip (quantitative, direction-only, pass FDR (q<0.05)); and an “Associations & papers over time” chart by publication year.
• Direction of effect: a donut of Increases / Decreases / No clear effect with the total in the centre and a legend with counts.
• Top exposures: a bar chart of the most-studied individual exposures.
• Top exposures by category and Top outcome phenotypes: bar charts ranked by association count.
• Effect on outcome: a bar chart of the net direction split across the corpus.
• Recent associations: a table of the newest records — Exposure · Outcome · Metric · Effect (95% CI) · p · Direction · Strength · Cohort · Source.

Data shown: Corpus counts; the quantitative / direction-only / FDR-passing split; associations and papers per year; direction of effect; per-exposure and per-category association counts; and per-record metric, estimate with 95% CI, p-value, direction, strength and cohort.

Key interactions

• KPI tiles and chart bars navigate: an exposure bar opens its profile; a category bar opens the Association Explorer pre-filtered to that category.
• “View all” buttons open a ranked-list modal of the full list.
• Click a row in Recent associations to open the association drawer.

Insights (Key findings)

Description: An auto-generated, filter-free summary of the corpus’s headline conclusions.

Purpose: See the pre-computed “so what” of the evidence base — which exposure families push outcomes up or down, the strongest statistically significant links, and the most replicated signals — without building a single query.

What you see

• Headline cards: Most outcome-increasing family · Most outcome-decreasing family · Densest evidence (the single hottest exposure-family → outcome-domain cell). Unclassified families are never reported as a headline.
• “What impacts what — and which way”: a diverging bar chart, one row per exposure family, extending left for decreases and right for increases, with counts on each side.
• “Strongest established links”: a ranked list of the FDR-passing, de-duplicated exposure→outcome pairs. Each row shows the exposure, a direction verb, the outcome, and a meta line (metric + estimate + CI, p, cohort · country, ✓ FDR) with an effect chip. The page discloses that the list is capped.
• “Most replicated & consistent”: pairs reported by two or more studies, ranked by a replication score; each row shows the number of studies and cohorts, the share agreeing on direction, and the FDR-passing count.

Data shown: Associations aggregated by exposure family (increase / decrease / null counts and net) and by exposure→outcome pair (study count, distinct cohorts, direction agreement, FDR-passing count, dominant direction); plus per-record metric, estimate, CI, p-value, FDR flag, cohort and country. A note on the page states that effect metrics are never pooled across studies here.

Key interactions

• “View all in the Explorer” under each list opens the Association Explorer — the strongest-links link pre-filters it to FDR-passing records.
• Click a diverging-bar row to open the Association Explorer filtered to that exposure family.
• Click a “strongest” row to open the association drawer. Clicking a “replicated” row opens a pair drawer listing every underlying study, each of which opens its own association drawer.

Browse

Find and inspect the evidence — associations, exposures and outcomes.

Association Explorer

Description: A full-width, filterable, sortable and paginated table of every curated exposure→outcome association in the corpus; clicking any row opens its full provenance drawer.

Purpose: The primary way to browse and narrow the whole evidence base — by exposure, outcome, statistics, cohort, geography, design, year and quality — and then trace any single result back to the exact place in its source paper.

What you see

• Filter bar: free-text search; Exposure category; Outcome phenotype; Direction (increases / decreases / no clear effect); Metric; Strength; Country; Study design; a Minimum-publication-year slider; and two toggles — Quantitative only and Passes FDR.
• Metric filter: grouped into Comparable metrics (OR, HR, RR, β, % change) and Study-specific — not comparable ones, plus a single option that selects every non-comparable metric at once.
• Result summary: total associations · quantitative vs direction-only · number passing FDR.
• Table columns: Exposure · Outcome · Effect (95% CI) · p · Significance · Direction · Replication · Model / subgroup · Strength · Cohort · Design · Source (year + journal). Every column is sortable, and records with a missing value always sort last, in either direction.
• Second row per record: a full-width sub-row carries the Metric, the exposure contrast (the amount of exposure the effect refers to), and — where the source paper worded them differently — the exposure and outcome as that paper reported them. A metric that cannot be compared across studies is marked “(not comparable)”.
• Significance column: passes FDR, nominal, ns, or not reported when the paper gives neither a confidence interval nor a p-value. An asterisk marks results whose source study reported no FDR correction.
• Replication column: how many independent cohorts report that exact pair and whether they agree on direction — teal = strong agreement, amber = partial, orange ⚠ = mixed; “1 cohort” when a pair is reported by a single cohort.

Data shown: Each row is one curated association: effect estimate and 95% CI, p-value and FDR outcome, direction, strength, cohort / country / sample size, model and adjustment set, journal, year and DOI. CSV export adds the association ID, CI bounds, FDR p-value, the direction as reported and the effect-on-outcome.

Key interactions

• Click a row to open the provenance drawer (see below).
• Sort, filter, export: click any header to sort; all filters apply live; Export CSV saves the currently filtered set; Reset filters clears everything; Prev / Next paginate.
• Deep links: arriving from the heatmap, Insights or a category tile pre-applies the matching filters, and the active filter is shown as a chip.

Exposures (directory)

Description: A searchable directory of every exposure that has at least one association in the corpus, with summary tiles and rows that link to each exposure’s profile page.

Purpose: Find an exposure and judge, at a glance, how heavily it is studied — association count, breadth of outcomes and cohorts, and FDR pass-rate — before opening its full profile.

What you see

• Summary tiles: Exposures · Families · Associations · Avg outcomes per exposure · With an ontology term.
• Controls: search box; Exposure-family filter; Ontology-term filter (With a term / Confirmed / Exact match / Suggested / No term yet); Sort by (Most associations / Most outcomes / Most cohorts / Name A–Z / % pass FDR).
• Columns: Exposure (name + sampling matrix) · Family · Ontology term · Associations · Outcomes linked · Cohorts · % pass FDR. Sortable and paged.
• Ontology term: the term's label, its ID (opens the term in its ontology) and its mark: Confirmed by a curator, Exact match (the name is the term's label or exact synonym) or Suggested (a close match awaiting a curator).

Data shown: One row per exposure that has at least one association: its association count, number of distinct outcomes and cohorts it appears with, and the share of its quantitative associations that pass FDR, plus its family and sampling matrix.

Key interactions

• Click a row to open that exposure’s profile page (not a drawer).
• Search / filter / sort: live; sortable headers stay in sync with the Sort-by menu; Prev / Next paginate.
• Search by ontology: the search also finds exposures whose ontology term, or a broader term of it, matches; such a row says which term it was found under (for example, organophosphate finds the pesticides mapped under that class).
• Ontology groups: a “Grouped under” chip on a profile opens this table filtered to every exposure under that broader term; clear the chip above the table to see all again.

Outcomes (directory)

Description: A searchable directory of every health outcome that has at least one association, organised by organ system, with rows that link to each outcome’s profile page.

Purpose: Find a health outcome and gauge how many exposures drive it, how many cohorts, and how much of the evidence passes FDR, before opening its full profile.

What you see

• Summary tiles: Outcomes · Organ systems · Associations · Avg exposures per outcome · With an ontology term.
• Controls: search box (also by ontology term and its broader terms); Organ-system filter; Ontology-term filter; Sort by (Most associations / Most exposures / Most cohorts / Name A–Z / % pass FDR).
• Columns: Outcome (name + type) · Organ system · Ontology term · Associations · Exposures linked · Cohorts · % pass FDR. Sortable and paged. The ontology term (MONDO, HP, OBA, EFO and others) is shown as on the Exposures directory.
• Outcome type is shown as binary, continuous, or mixed when the underlying papers reported it both ways; hovering says whether the type was declared by the paper or inferred from the metrics actually reported.

Data shown: One row per outcome that has at least one association: its association count, number of distinct exposures and cohorts, share passing FDR, plus its organ-system category and type.

Key interactions

• Click a row to open that outcome’s profile page.
• Search / filter / sort: live; sortable headers; Prev / Next paginate.

The association drawer

The shared panel that makes every number traceable.

Description: The side panel that opens whenever you click a single association anywhere in the application — the central artefact for traceability and the reason every number in the console is auditable.

Purpose: Turn one row into a credibility judgement: what the effect is, whether it replicates across cohorts, how it was modelled, and exactly where in which paper it came from.

What you see

• A plain-English effect sentence, significance-gated — a non-significant estimate reads “No clear association”, never a fabricated direction.
• Status badges: direction, quantitative vs direction-only, metric, replication and passes-FDR.
• “Cross-cohort evidence”: a random-effects (DerSimonian–Laird) forest plot — every cohort’s estimate and 95% CI, a pooled diamond, I² heterogeneity with a low / moderate / high label, the share of cohorts agreeing on direction, and a line stating whether the pooled models used a consistent adjustment set.
• Detail list: effect size, p-value and FDR p-value, strength, exposure contrast, linked exposure and outcome, cohort / country / n, study design, analysis n, model, adjusted-for covariates, and evidence grade.
• Data-quality note: where a source paper reported a value that cannot be true for its metric (for example an odds ratio of zero, or a confidence interval that excludes its own estimate), the console withholds the impossible number and says so here, rather than silently reinterpreting it.
• Provenance block: the source paper (with Open PDF), journal · year, DOI, the exact page or table reference, and links to any published raw-data assets.
• Embedded PDF viewer: opens the source paper with the value outlined where it is printed. A label says how it was found: Exact value (the estimate and its CI as printed, in a table row or a sentence), Quoted text, Table or figure (the value is in a figure image or could not be matched, so the caption of the cited table or figure is outlined) or Cited page.

Data shown: Everything recorded for a single association, joined to its cohort, study, model specification, adjustment set, quality grade and provenance record — plus every other cohort reporting the same exposure→outcome pair, for the cross-cohort pooling.

Key interactions

• Click a cohort row in the forest to jump to that cohort’s association.
• Show in paper reopens the PDF at the outlined value; All curated values in this paper opens the Paper evidence page.
• Exposure / outcome names open their profile pages.

Analyze

Quantify, pool and summarise the evidence.

Exposome Atlas

Description: A single page that assembles the signature views of a systematic exposome-wide analysis over the whole corpus — the association landscape, where the evidence concentrates, cross-cohort replication and reproducibility, the exposome globe, and how densely connected the exposome is.

Purpose: A one-screen overview of the corpus the way an exposome-wide study is usually presented — for orienting a new user, briefing a stakeholder, or spotting at a glance which exposures and phenotypes carry the most (and the strongest, most replicated) evidence.

What you see

• Headline counts: Associations · Exposures · Phenotypes · Cohorts · Countries · Pass-FDR.
• Exposome-wide association landscape (Manhattan / ExWAS): every quantitative test ranked by significance and grouped by exposure family, with a dashed nominal-significance line and −log₁₀(p) capped for readability. A phenotype picker focuses it on one outcome (a full-shape ExWAS) or shows the most-significant skyline across all phenotypes.
• Exposome load by family / Phenome load by organ system: two bar charts of where the associations concentrate.
• Phenome–Exposome Atlas matrix: the broadest-footprint exposures × phenotype domains, with a Direction / Volume / % FDR toggle; grey cells are not studied. Click a cell to open the exposure.
• Replication & heterogeneity across cohorts: a random-effects (DerSimonian–Laird) forest of the most-replicated pairs, with median-I² and direction-agreement KPIs.
• Cross-cohort concordance: a log–log scatter pairing the effect estimate from two independent cohorts of the same pair against a y = x diagonal — points on the diagonal reproduce. Headline r and % same-direction.
• The exposome globe: a circular chord diagram with two modes. Measured (NHANES), the default, joins exposure biomarkers by their participant-level partial correlation in the NHANES survey (adjusted for age, sex, race/ethnicity, income and education; chords for |r| > 0.25, blue positive and orange inverse). Shared-effect joins exposures that act on the same health outcomes in the corpus. Nodes are coloured by family, with a link to the interactive Co-exposure network.
• How connected is the exposome?: a histogram of shared-effect link strength across exposure pairs.

Data shown: All panels except the measured globe are computed from the corpus’ associations (curated and NHANES atlas). The measured globe comes from public NHANES participant-level data; the shared-effect globe shows exposures acting on the same outcomes, not correlation of measured exposure levels. Per-exposure variance explained (R²) is not shown, because it needs subject-level data the corpus does not hold.

Key interactions

• Pick a phenotype to refocus the Manhattan landscape.
• Toggle the matrix mode (Direction / Volume / % FDR); click a cell to open that exposure.
• Click a Manhattan point for its provenance drawer; open the interactive network from the globe.

Pathway Analyst

Description: Pick an exposure and an outcome to see how they are related: the direct evidence, the indirect pathways the corpus supports, and a written synthesis grounded strictly in that evidence.

Purpose: Answer “how might X affect Y?” in one place — what the studies report directly, and which shared co-exposures or mechanisms could connect them — with every statement traceable to a study.

What you see

• Pickers: an exposure (only exposures with data are listed) and an outcome; once an exposure is chosen the outcome list shows the outcomes directly linked to it, and typing any other outcome runs an indirect-only analysis. Suggested pairs appear as chips.
• Direct relationship: association and cohort counts, the number passing FDR and the overall direction, then each reported estimate with its 95% CI and a link to the study, and the curated mechanisms for the pair.
• Indirect pathways: hypothesised links the corpus structure supports — a co-exposure measured alongside the exposure in many studies and itself linked to the outcome, or a shared mechanism (a mediator the exposure shares with another exposure linked to the outcome) — each with its evidence and study.
• AI synthesis: a short narrative written by the on-premises language model from the evidence above only, citing the paper of every claim. If the model is busy, the evidence panels still show and the synthesis says so.

Data shown: curated associations, mechanisms and co-measurement for the chosen pair. The synthesis adds no external data and no citations of its own.

Key interactions

• Click Analyze or a suggested pair; open any cited study from its link.

Effect sizes

Description: A single card that switches between publication-style statistical charts — Volcano, Manhattan / ExWAS, Forest, and Meta-analysis — computed over the quantitative associations.

Purpose: Inspect the numeric evidence — effect size vs significance, a per-outcome ranking of exposures, per-exposure confidence intervals, and random-effects pooling of replicated pairs — while never mixing incompatible effect metrics (ratio metrics vs β / % change are kept apart).

What you see

• View switch: Volcano · Manhattan / ExWAS · Forest · Meta-analysis.
• Filters: Outcome; Effect metric (OR / HR / RR / β / % change); Exposure category; a “replicated pair” selector for meta-analysis; and a Highlight-FDR-passing toggle (controls show or hide per view).
• Volcano: effect size (x, log-scaled for ratio metrics) vs −log₁₀(p) (y), with an FDR significance line; points coloured by direction, faded when not FDR-passing.
• Manhattan / ExWAS: every tested exposure ranked by significance, grouped and coloured by exposure category, with a dashed FDR line.
• Forest: the leading exposures for the chosen outcome, each row an estimate + 95% CI with cohort / country / n. Where a study never reported the cohort’s sample size, the row shows “—” rather than a fabricated zero.
• Meta-analysis: a random-effects (DerSimonian–Laird) pooled forest for one replicated pair — one row per cohort plus a pooled diamond labelled with the number of cohorts and I².
• Stat cards: Quantitative tests · Pass FDR (q<0.05) · Strongest link · Most significant — and in Meta view: Pooled estimate, Heterogeneity I², Cohorts pooled, Direction agreement.

Data shown: Per association: estimate + 95% CI, metric, direction, p-value, FDR flag, cohort / country / n / year. Pooled outputs: random-effects estimate + CI, I² with a low / moderate / high label, τ², number of cohorts, and direction agreement.

Key interactions

• Switch views or change filters to redraw the chart and stat cards.
• Click a volcano or Manhattan point, or a forest row, to open that association’s drawer.

Correlation (Exposure × Outcome matrix)

Description: A matrix view with a coverage card, a dedicated research-gaps panel, and two heatmaps — exposure categories × outcome phenotypes, and the most-studied individual exposures × phenotypes — recoloured by a Count / Net-direction / % FDR toggle.

Purpose: See how exposure families map onto outcome domains across the whole corpus, and exactly where the empty intersections (research gaps) are — with click-through to the matching association list or exposure profile.

What you see

• Mode toggle: Count · Net direction · % FDR-passing.
• Coverage card: KPIs — Category cells, Studied, Empty (gaps), Coverage %.
• “Highest-value research gaps”: its own panel, with a count KPI and a numbered list of the gap pairs where both the exposure family and the outcome domain are well studied, yet their intersection holds no associations at all. The panel discloses how many gaps are listed out of the total found.
• Primary heatmap: exposure categories (rows) × outcome phenotypes (columns).
• Secondary heatmap: the most-studied individual exposures × outcome phenotypes.
• Notes: each heatmap states how many cells are filled vs empty.

Data shown: Per cell: association count, net direction (increases minus decreases), and the share of quantitative associations passing FDR. Colour encoding: teal sequential (count), blue sequential (% FDR), and a diverging blue→grey→orange scale (net direction). Gap ranking scores each empty cell by row-total × column-total.

Key interactions

• Toggle mode to recolour both heatmaps and their colour bars.
• Click a category cell to open the Association Explorer filtered to that category pair.
• Click an individual-exposure cell to open that exposure’s profile.

Mechanisms & pathways

Description: A Sankey diagram of exposure family → mechanism theme → outcome domain, followed by a filterable grid of mechanism-of-action cards.

Purpose: Trace the biological routes linking exposures to outcomes and use the Pathway + Outcome-domain filters to reveal which exposures act on an outcome through the same mechanism.

What you see

• Sankey: node columns for exposure category, canonical pathway theme and outcome domain, with link width proportional to the number of curated mechanisms.
• Filters: Pathway (canonical) · Outcome domain · Mechanism type · Evidence source (Paper-reported / Inferred / Literature), plus a live count of matching mechanisms.
• Mechanism cards: each shows a pathway-theme tag, the exposure→outcome pair (both linked to their profiles), and Pathway, Mediators, Direction (▲ upregulates / ▼ downregulates), plus its mechanism ID. The grid is paged.

Data shown: Mechanism-of-action records: pathway theme, reported type, pathway, mediators, direction, evidence source, and the linked exposure and outcome.

Key interactions

• Change any filter to re-filter the cards and update the count (the Sankey is drawn once).
• Click an exposure or outcome name to open its profile page.

Explore

Explore the structure and geography of the evidence.

Network (Exposome network)

Description: A bipartite graph with exposures in a left column and health outcomes in a right column, joined by curved edges.

Purpose: See which exposures connect to which outcomes across the literature and in what direction, then jump into any profile. Only the most-connected nodes are drawn unless a category is chosen.

What you see

• Filters: Exposure category · Outcome phenotype · Minimum studies per link (slider) · FDR-passing links only (toggle).
• Graph encoding: node size = degree (number of linked studies); node colour = its category; edge colour = the pair’s dominant direction; edge thickness = the number of studies behind that pair.
• Header and note: live counts of exposures, outcomes and links; a legend; and a note that only direction-bearing links are shown and that each edge aggregates all studies for that pair.

Data shown: Exposure and outcome nodes with their degree and category colour; edges aggregated per exposure→outcome pair with a study count and a dominant direction. Associations with no clear effect are excluded; the FDR toggle restricts to FDR-passing evidence.

Key interactions

• Hover a node to isolate its edges and show its name and linked-study count. Hovering an edge shows the pair, its study count and its direction.
• Click a node to open that exposure’s or outcome’s profile.
• Filters and the FDR toggle redraw the graph live.

Co-exposure (interactive exposome network)

Description: A live, force-directed network of exposures linked to one another by shared evidence, under one of two selectable link layers.

Purpose: Explore how exposures relate to each other — because they act on the same outcomes, or because the same study measured both — by dragging, zooming, pinning and searching nodes.

What you see

• Link-layer toggle: “Shared health effects” — exposures linked when they act on the same outcomes (green = concordant direction, grey = mixed; thicker = more shared outcomes); and “Co-studied” — exposures linked when the same study measured both (thicker = more shared studies).
• Controls: Exposure category · Minimum link strength · Maximum exposures · Show shared outcomes · Find node (with autocomplete) · Re-run layout · Reset view.
• Graph: exposures as circles sized by association count and coloured by family; optional outcome nodes as small squares. A hint pill explains drag / double-click / scroll / click, alongside a legend and live counts.

Data shown: Exposures ranked by association count (capped by the Max-exposures control), with two independently computed edge layers: shared outcomes (including a directional-concordance test) and shared source studies.

Key interactions

• Direct manipulation: drag a node to move it; double-click to pin or unpin it; scroll to zoom; drag the background to pan; hover to highlight a node’s neighbourhood.
• Click a node to open its profile. “Find node” searches, highlights and recentres on a node.
• “Re-run layout” / “Reset view” reheat the simulation or restore the original pan and zoom.

Geography & cohorts

Description: A proportional-symbol world map of the contributing cohorts by country, with regional summary charts and a top-countries table.

Purpose: See where the evidence base’s cohorts come from, how coverage breaks down by region, study design and sample size, and drill from any country into its associations.

What you see

• Map: one bubble per country — bubble size = number of contributing cohorts, bubble colour = world region (Europe, North America, Asia, Latin America, Oceania, Africa, Other). Well-represented countries are labelled with their ISO code. A note under the map discloses any cohorts whose country cannot be plotted.
• Cohorts by region: a bar chart of cohort counts per region.
• Study designs: a bar chart of cohort counts per study-design type.
• Sample-size distribution: a box plot of cohort sample size per region.
• Top contributing countries: a paged table — Country · Region · Cohorts · Associations · Total sample (n) · Designs. Where sample size was never reported, the total reads “—” and the number of cohorts that did report one is shown alongside, so a partial total is never mistaken for a complete one.

Data shown: Cohorts aggregated by country — cohort count, summed sample size, distinct study designs and the number of associations contributed — plus each cohort’s region and total sample size.

Key interactions

• Hover a bubble for country, cohort and association counts, total sample size and number of designs.
• Click a bubble or a table row to open the Association Explorer filtered to that country.

Entity profiles

The per-exposure and per-outcome deep-dive screens.

Exposure profile

Description: An entity-specific dashboard for a single exposure, reached by clicking that exposure anywhere in the application.

Purpose: The one-exposure, outcome-wide (reverse-ExWAS) view: every outcome linked to this exposure, with direction, effect size and significance, its biological mechanisms, and how it is measured across studies.

What you see

• Header and KPI tiles: name, family, sampling matrix and counts; then Associations · Linked outcomes · Quantitative · Pass FDR · Mechanisms · Cohorts.
• Ontology term: the exposure's term with its ID (opens the term in its ontology), its mark (Confirmed, Exact match or Suggested), its definition, the cleaned name it was matched on, and Grouped under: the broader terms it shares with other exposures, each with their number. Curators and administrators can confirm or reject the term here, or choose another on the Ontology & FAIR page.
• “Outcome-wide scan · outcomes ranked by significance”: a Manhattan (reverse-ExWAS) plot — every outcome tested against this exposure, grouped and coloured by organ system, y = −log₁₀(p), with a dashed FDR line; points above the line survive FDR correction.
• “Effect profile” volcano (effect estimate vs significance) and a “Direction of effect” donut.
• “Linked outcomes” bar chart and a scrollable “Mechanisms of action” list.
• “Outcome footprint by organ system” bar chart, and “Measurement across studies” — the sampling matrix, unit and assay method reported by each contributing paper.
• The associations table with a Significance / Effect size / Quality rank toggle: Outcome · Metric · Effect (95% CI) · p · Direction · Strength · Quality · Cohort · Year. Paged.
• An exposure with no associations still opens, showing an explicit empty state rather than a broken page.

Data shown: The exposure and all of its associations (outcome, metric, estimate + CI, p-value, direction, strength, cohort, country, year, FDR status), its curated mechanisms, and its per-paper measurement variants.

Key interactions

• Click any Manhattan or volcano point, or any table row, to open the association drawer.
• Click a bar to open the outcome profile or the filtered Association Explorer.
• Rank toggle re-sorts the associations table by significance, effect size or evidence quality.

Outcome profile

Description: An entity-specific dashboard for a single health outcome.

Purpose: The one-outcome ExWAS view: every exposure associated with it, their direction, size and significance, the convergent biological pathways, and how the outcome is measured.

What you see

• Header and KPI tiles: name, organ system, type, and counts; then Associations · Linked exposures · Quantitative · Pass FDR · Mechanisms · Cohorts.
• Ontology term: the outcome's phenotype term (MONDO, HP, OBA, EFO and others) shown and reviewed as on the exposure profile, with the broader terms it shares with other outcomes.
• “ExWAS · exposures ranked by significance”: a Manhattan plot of every exposure tested against this outcome, grouped and coloured by exposure family, with a dashed FDR line.
• “Top exposures” bar chart and a “Direction of effect” donut.
• “Mechanisms of action” list and an “Evidence strength” bar chart (Strong / Moderate / Weak / None).
• “Convergent mechanisms · exposures grouped by shared pathway”: exposures grouped by pathway theme, each group showing how many exposures converge via that pathway, as clickable chips.
• “Exposure sources by category” bar chart, and “Measurement across studies” — the measure, assay and matrix reported by each contributing paper.
• The associations table with the same rank toggle: Exposure · Metric · Effect (95% CI) · p · Direction · Strength · Quality · Cohort · Year. Paged.

Data shown: The outcome and all of its associations, its curated mechanisms grouped by pathway theme, and its per-paper measurement variants.

Key interactions

• Click any Manhattan point or table row to open the association drawer.
• Click a bar or a pathway chip to open the exposure profile or the filtered Association Explorer.
• Rank toggle re-sorts the associations table.

Master Data

The underlying master tables and exports.

Description: A tabbed browser over the underlying master tables, each with its own search, filters, sorting, pagination and CSV export.

Purpose: Inspect the raw curated records table by table, curate the exposure classification, and export any filtered and sorted view for offline analysis.

What you see

• Studies & papers: Paper ID · Title · Journal · Year · Study design · Exposures (curated) · Screened (study) · Associations.
• Cohorts: Cohort ID · Acronym · Country · Region · Study design · Sample (n) · Mean age · Paper.
• Exposures: Exposure ID · Name · Category · Matrix · Unit · Associations.
• Outcomes: Outcome ID · Name · Phenotype · Type (binary / continuous / mixed) · Molecular marker · Associations.
• Mechanisms: Mechanism ID · Exposure · Outcome · Type · Pathway · Mediators · Direction · Evidence.
• Exposure classification: Exposure ID · Exposure · Category · Classification · Associations. The Category column is a dropdown: changing it re-files the exposure everywhere, the Classification column then reads curated with who changed it and when, and a curated row can be reset to the automatic value.
• Provenance: Provenance ID · Association · Exposure → Outcome · Paper · Reference.
• Every tab has a search box, its own dropdown filters, sortable headers, a result count and a pager.

Data shown: Papers, cohorts, canonical (harmonised) exposures, canonical outcomes, mechanism records, the exposure classification, and the per-association provenance table that links every record to its exact source reference.

Key interactions

• Switch tabs, search, filter, sort, paginate.
• Click a row in the Exposures or Outcomes tab to open that entity’s profile.
• Change a category in the Exposure classification tab (curators and administrators); reset it to the automatic value at any time.
• Export CSV downloads the current filtered and sorted view of the active tab.

Curation and administration

Every account has one of three roles. A researcher reads the corpus and every analysis. A curator also curates papers (Curate a paper), reviews ontology mappings (Ontology & FAIR) and edits the master data (Exposure Category Master, and the exposure classification under Master Data). An administrator also manages accounts and roles (Users) and sees the sign-in activity. Self-registered accounts start as researchers; an administrator grants the curator or administrator role. The menu shows each role only its own screens, and the server refuses the rest.

Paper evidence

Description: One paper's curated values beside its PDF (from the association drawer: All curated values in this paper).

What you see

• KPI tiles: curated values, and how many were found as their exact value, in a quoted sentence, only by their table or figure, or not located.
• Curated values: each value with its exposure → outcome, subgroup, effect and CI, how it was found and the page / table reference; filter by how it was found, or search.
• Paper: the PDF with every located value outlined (teal: exact value; blue: quoted text; amber dashed: table or figure caption) and the chosen one in orange. Zoom with − and +.

Key interactions

• Choose a value to scroll the PDF to it; click an outline on the PDF to choose its value.

Curate a paper

Description: Add a new paper to the corpus. Upload its PDF; the on-premises extraction engine drafts the schema records (associations, exposures, outcomes, cohort, adjustment set, model and mechanisms); a curator reviews each part beside the PDF and approves the paper into the database. All processing is local.

What you see

• Tiles: papers in progress, ready for review (with the results still to check), approved (and how many still have results not yet verified), and not curated.
• Batch and uploads: a paper a curator uploads here is reviewed beside its PDF before approval. Papers curated in batch by the engine (python -m scripts.curation.curate_batch) join the corpus after the system checks and show System-checked; they are not a review backlog, though a curator can still open and review one. The Uploads / Batch filter separates the two.
• Papers: every upload with its title, who uploaded it and when, its status, its number of results, how far its review has got, and what the checks found. Filter by status or search titles and file names.
• Upload papers opens a dialog: drop one or several PDFs (or drop them anywhere on the page); each uploads with its own progress bar and is curated at once, showing its progress in the list.
• + Supplementary data adds results printed only in supplementary files. Download the Excel template (a read-me, then Study, Associations, Mechanisms and Covariates & model sheets, with dropdowns), fill it in and upload it, on its own (it becomes a paper without a PDF) or attached to a paper uploaded here (its rows join that paper; rows it already has are not added twice). The workbook is checked at once: every problem is listed with its sheet, row and column; errors must be corrected and the workbook uploaded again, warnings do not stop it. Nothing is saved until Add for review. Workbook rows are marked From the workbook in the review and load with their provenance on approval. In a paper's workspace, + Supplementary data attaches a workbook to that paper.
• Live progress: while the engine works, the paper shows each step (reading the PDF; title, journal and design; results tables page by page; results text; completeness check; consistency checks and exposure families; methods, adjustment set and mechanisms) with its latest message. The page can be left; the paper keeps curating.

One paper, section by section

• Overview: what the paper found (results; significant increases and decreases; no clear effect; mechanisms; pairs new to the corpus), the study (design, cohort, place, participants, recruitment, follow-up, age, sex, model, adjustment set), a curation checklist, a forest plot of every result with a confidence interval (one panel for ratio measures on a log scale, one for differences; choose a result to check it) and the quality checks.
• Associations (beside the PDF): each result with its exposure → outcome, subgroup, contrast, effect and CI, whether it is significant, and whether its numbers were found in the paper and where. Choosing a result outlines it in the PDF. Each result is accepted, corrected or rejected. Accept all found in the PDF accepts every result whose estimate and CI are printed exactly as drafted; Add a result adds one the engine missed. Results show at once while their evidence is located, and the workspace opens on the next result to review. Keyboard: J/K or the arrows move, A accepts and moves to the next result to review, R rejects, E corrects, X selects (then accept or reject the selection together), U undoes the last decision, N jumps to the next result to review, ? lists the shortcuts. Every decision is saved as it is made. After approval, Next paper to review opens the next paper waiting.
• Exposures and Outcomes: each one with the corpus entity it is harmonised to, its family or domain, what this paper found, how many other papers study it and which outcomes (or exposures) they link it to. Choose one to review its results.
• Mechanisms & pathways (beside the PDF): the biological pathways the Discussion proposes, each as exposure → pathway → outcome with its mechanism type, mediators, direction, confidence (high: shown with the paper's own data; moderate: argued from cited evidence; low: speculation) and the sentence that states it, outlined in the PDF. A mechanism is drafted only when its sentence is found in the paper. Accept, correct, reject or add one.
• Methods & cohort (beside the PDF): the study and cohort details, the adjustment set (the standard covariates flagged, the list as printed and the sentence that gives it) and the model (method, multiple testing, sensitivity analyses, missing data). Numbers are kept only when the paper prints them. Each part is marked checked.
• Network: exposures, the proposed pathways and outcomes; associations are coloured by the direction of their significant results and mechanisms are dashed through their pathway. Choose a link or a node to review it.
• In the corpus: each exposure–outcome pair against the other curated papers: reported before (and whether in the same direction), related evidence only (the same exposure with other outcomes of the same domain, which is often the same condition curated under another name, never counted as a replication) or new. Choose a pair for the other papers' estimates. Below: the NHANES-measured correlations between the paper's exposures, and the pathways other papers give for them.
• Processing: every step the engine took, with times.

How the draft is made: the engine reads the paper's opening text for its metadata; finds the pages that hold results (tables, figures, forest plots) and reads up to eight of them as images; reads the Results section (not the Discussion, so effects quoted from other studies are not picked up); and merges the rows on their numbers. A completeness check then lists every effect with a confidence interval printed in the paper and adds those the draft lacks, with numbers taken from the page. Finally it reads the Methods for the cohort, the adjustment set and the model, and the Discussion for the mechanisms the authors propose. Numbers are taken only from the paper, and the direction of each row is the sign of its estimate.

Key interactions

• Every decision is saved with the paper's draft as it is made.
• Approve into the corpus loads the paper: rejected results and mechanisms are left out, accepted and corrected results load as verified by the curator, results not reviewed load unverified. The PDF is published with the paper, the exposure and outcome harmonisation is rebuilt, and every screen refreshes.
• An approved paper stays open for review. Changes are marked Changes not in the corpus yet until Update the corpus reloads the paper with them. Discard removes an upload that is not approved.

Ontology & FAIR

Description: The mapping of every exposure and outcome to public ontologies, its curator review, and the corpus as FAIR linked data.

What you see

• FAIR dataset: the published JSON-LD export of the whole corpus (CC BY 4.0) with its download links.
• Ontology mappings: each entity with its type, mapped term, similarity, the cleaned name it was matched on, and status, filterable by All / Suggested, to review / Exact match / Confirmed / Rejected / Unmapped and searchable by entity or term. An exposure or phenotype page links here for its own mapping.
• Decide many at once (curators and administrators): accept the suggestions in view whose similarity is at least a chosen value, or reject those below one. A confirmation lists them first, and every decision goes to the audit trail.
• Review: a curator or administrator can accept a mapping, reject it, reassign it to another term (by CURIE or IRI), or return it to the review queue, with a note.
• FAIR JSON-LD: the selected entity, and a sample association, as linked data.
• Provenance & audit trail: every automated mapping and curator decision, newest first, with who made it and the note; filter to curator decisions only.

Key interactions

• Click a row to see its JSON-LD; use the review actions on rows that need a decision.

Email settings

Description: The mail server the console sends its notifications through, the sender, and the administrators who are told about new registrations and curated papers (Admin → Email settings).

What you see

• Status: whether email is on, and who changed the settings last. These settings are the only ones the console uses: until they are saved here, no email is sent, and switching email off here stops all of it.
• Mail server: the on/off switch, server, security (STARTTLS, SSL/TLS or none), port, username and password, the sender, and the administrator addresses, one per line.
• Send a test message: sends one message at once with the saved settings and says whether the server accepted it, or why not (wrong password, unreachable server, email switched off).

Key interactions

• Password: the saved password is never shown. Leave the field empty to keep it, type a new one to replace it, or choose Remove it. It is stored encrypted with the server's own key, so a copy of the database on another server cannot use it: enter it again there.
• Save settings applies to the next message. Every change is written to the activity log with the fields that changed (never the password), and every message, sent or not, is listed under Sign-in activity → Email.

Users

Description: Every account on the console (Admin → Users, or the Admin link next to your name).

What you see

• KPI tiles: Accounts · Active · Awaiting approval · Administrators · Disabled or rejected. Click a tile to filter the table to it.
• Awaiting approval: researchers who registered and cannot sign in yet, oldest first, with Approve and Reject. An approved researcher is sent an email and can sign in at once.
• The table: user (name and email), institution, role (Administrator, Curator or Researcher), status, registration date and last sign-in, with search and Role and Status filters.

Key interactions

• + Add user creates an account directly: full name, email, optional username (the email by default), password and role.
• The ⋮ menu on a row: Make researcher / Make curator / Make administrator (whichever the account is not), Reset password, and Disable / Enable account. Disabling an account or resetting its password ends that user's sessions. You cannot disable or demote yourself if you are the last administrator.
• Export CSV downloads the filtered table.

Sign-in activity

Description: The activity log (Admin → Sign-in activity).

What you see

• KPI tiles: sign-ins and failed attempts over 24 hours, sign-ins and users signed in over 7 days, and the last sign-in.
• Sign-ins per day for the last 14 days.
• The log: time, user, event, IP address, client (browser and system) and details. Choose Sign-ins, Accounts (registrations, approvals, role changes, password resets), Data changes (taxonomy edits), Email (each notification sent, failed or not sent) or All; narrow by outcome and period, and search by user, IP or details.
• Repeated failed sign-ins lock the account or address for a while; those attempts appear as Blocked.

Exposure Category Master

Description: The list of exposure families the whole corpus is reported against — those the keyword classifier knows about, and any a curator has added.

Purpose: Extend and correct the taxonomy itself. This is the one screen in the application that creates new data rather than reading it, which is why it is separate and why it is restricted.

What you see

• KPI tiles: Categories · User-added · Empty (nothing filed yet) · Still in “Other”.
• The table: Category (with its colour) · Key · Colour · Origin (system or user-added) · Status (active or retired) · Exposures · Associations.
• Filters: search, Origin, Status, and Holding (in use / empty).

Key interactions

• + Add category opens a dialog: a name, and a colour chosen from a swatch grid in which the colours already taken by another family are greyed out. The new family is selectable immediately — in the Exposure classification dropdown and in every category filter, chart legend, heatmap row and network node. That is what lets an exposure stranded in “Other” be filed somewhere meaningful even when no suitable family existed before.
• Edit renames or recolours a family; the change reaches every chart and legend at once.
• Retire removes a system category from the dropdowns while keeping its colour, so anything filed under it historically still renders. It can be restored later.
• Delete removes a user-added family outright. System categories cannot be deleted — the classifier still refers to them, so one would reappear on the next rebuild.
• A family that still holds exposures cannot be retired or deleted: those exposures would land in a category nobody can select. Re-file them first, in Master Data → Exposure classification.

Glossary

Terms used consistently across the screens.