Coral Epigenetic Network Explorer

An interactive, static web resource that lets readers explore the predicted multi-layered epigenetic regulatory networks described in the manuscript “Multi-Layered Epigenetic Regulation in Three Reef-Building Corals” (Acropora pulchra, Porites evermanni, Pocillopora tuahiniensis).

It is a publication companion to the analysis repository urol-e5/deep-dive-expression: readers can move from a plain-language overview to detailed inspection of individual miRNAs, lncRNAs, genes, predicted interactions, DNA methylation, and cross-species differences — and download the underlying evidence.

Scientific framing. Every interaction shown is computationally predicted from sequence complementarity (miRanda) and/or expression coexpression (Pearson correlation, n = 5 per species, unadjusted p-values). These are hypotheses for future experimental testing — not validated causal relationships. See docs/SCIENTIFIC_INTERPRETATION.md.

Scientific context

The study integrates matched RNA-seq, small RNA-seq, and whole-genome bisulfite sequencing to characterize DNA methylation, miRNAs, and lncRNAs across three coral species and — for the first time in cnidarians — describes predicted epi-miRNAs (miRNAs targeting epigenetic machinery) and competing endogenous RNA (ceRNA) networks (lncRNAs predicted to sequester miRNAs).

Features / views

View	What it shows
Home	Overview, multilayer diagram, entry points, caveats.
Network Explorer	Cytoscape.js graph of predicted miRNA / lncRNA / mRNA / epi-machinery interactions with rich filters, presets, node/edge detail, and PNG/SVG/CSV export.
Regulatory Story	Curated chains (lncRNA sponge → miRNA → mRNA → function), incl. the ptuh-mir-novel-4 → TNRC6 ceRNA example.
Compare Species	Cross-species cards, comparative charts, and the miR-100 panel.
Epigenetic Machinery	Epi-miRNA targets by functional category (TET3, MBD, AGO, TNRC6, …).
Methylation Landscape	Global CpG methylation and feature-level context.
Evidence Table	Searchable / sortable / exportable table of every displayed interaction.
Methods & Caveats	Plain-language methods and the limitations that must frame interpretation.
Downloads	Grouped datasets with provenance + the data manifest.

Tech stack

• Astro static site + TypeScript
• Cytoscape.js for network visualization
• Inline, accessible SVG charts (no heavy charting dependency)
• Node preprocessing pipeline (csv-parse) → browser-ready JSON
• Vitest tests · GitHub Actions → GitHub Pages

No backend or database. Everything is prebuilt to static JSON in public/data/.

Local setup

npm install        # install dependencies
npm run dev        # build web data, then start the dev server
# → http://localhost:4321/CENE/

Development commands

Command	Purpose
npm run dev	Build data + start dev server.
npm run build:data	Regenerate public/data/*.json from data/source/.
npm run validate:data	Validate the generated data (fails on errors).
npm run fetch:data	Re-download source CSVs from deep-dive-expression.
npm test	Build data + run the Vitest suite.
npm run build	Production build (runs build:data first) → dist/.
npm run preview	Serve the production build locally.

Data preprocessing

Source CSVs are vendored under data/source/ (verbatim copies of deep-dive-expression outputs — see data/source/README.md). The pipeline in scripts/build_web_data/ reads them, normalizes column names while retaining original identifiers, validates required columns, reports (never silently drops) missing fields, and writes:

• public/data/network/{Apul,Peve,Ptuh}.json — per-species nodes + edges
• public/data/{summary,epimachinery,methylation,mir100,regulatory-stories,downloads}.json
• public/data/data-manifest.json, build-summary.json, dictionaries/network.json

npm run build:data      # regenerate everything
npm run validate:data   # then validate

Build & GitHub Pages deployment

npm run build outputs a fully static site to dist/. Deployment is automated by .github/workflows/deploy.yml: on push to main it installs, builds data, validates, tests, builds the site (with the correct repo subpath), and deploys to GitHub Pages. See docs/DEPLOYMENT.md.

The subpath is configurable via BASE_PATH (default /CENE) so forks and user-pages deployments work without code changes.

Directory structure

CENE/
├── data/source/              # vendored source CSVs (verbatim, do not edit)
├── scripts/build_web_data/   # preprocessing pipeline (CSV → JSON) + validation
├── public/
│   ├── data/                 # generated JSON (git-ignored; rebuilt by build:data)
│   └── favicon.svg
├── src/
│   ├── components/           # BarChart.astro, …
│   ├── layouts/Base.astro
│   ├── lib/                  # site config, data reader, pure filter helpers
│   ├── pages/                # one .astro per view
│   ├── scripts/              # client TS (network.ts, evidence.ts)
│   └── styles/global.css     # design system
├── tests/                    # Vitest (filters + generated-data integrity)
├── docs/                     # DATA_SOURCES, SCIENTIFIC_INTERPRETATION, DEPLOYMENT, ADDING_DATA
└── .github/workflows/deploy.yml

Data provenance

Every dataset maps to an original deep-dive-expression file. See docs/DATA_SOURCES.md and the generated public/data/data-manifest.json.

Adding new datasets

See docs/ADDING_DATA.md.

Known limitations

• Predicted interactions only (no experimental validation); n = 5 per species; unadjusted p-values.
• A machine-readable per-feature methylation summary is not available in the source repo (only a figure); the Methylation view shows global values + manuscript-derived qualitative levels and labels the numeric per-feature table as awaiting data.
• Non-epigenetic-machinery genes are shown by transcript ID; functional annotation is currently only joined for epi-machinery targets.
• Cross-species comparisons are affected by differing genome-assembly quality — interpret cautiously.

License / attribution

Scientific data © the study authors, from urol-e5/deep-dive-expression. This resource reproduces their published statistics and framing; consult the manuscript and analysis repository as the source of truth.