EBV Genome Atlas
A curated, interactive database of Epstein–Barr virus genomic variation across EBV-associated diseases, built on whole-genome analysis with machine-learning and tri-rheostat (immune evasion / latency / lytic) annotations.
Sample composition
Geographic distribution of available genomes
Worldwide geographic origin of the 1,202 EBV whole genomes (color depth = sample count). This map reflects data availability and sampling, not disease prevalence.
Genome overview
EBV genome (NC_007605.1, 171,823 bp). Top: gene annotation coloured by tri-rheostat class. Bottom: mutation frequency across the genome.
Browse
Browse by disease
Select a disease to see its tri-rheostat recipe and top model-prioritized candidate positions.
Browse by region
Select a geographic region to browse its samples.
Advanced Search
Combine filters to search variants by gene, axis, impact, functional class and selection.
Example queries
Click an example to run it in the Browse page.
Genes
BCRF1 (vIL-10, immune evasion) LMP-1 (latency signalling) BHLF1 (lytic activity) EBNA-1 (nuclear antigen)Tri-rheostat axes
Immune evasion Latency signalling Lytic activityDiseases
NPC GC BL CAEBVPositions
8873 (BCRF1) 171708 (LMP-1) 40278 (BHLF1)Genome Browser
Multi-track linear browser — genes (coloured by functional class, arrows show transcription direction), mutation frequency, tri-rheostat positions (▲ immune, ● latent, ◆ lytic), and dN/dS selection (red ω>1, blue ω<0.5, gray neutral). Scroll to zoom; drag to pan; hover for details.
Variant frequency heatmap
Variant lookup
Look up a genomic position to see its allele, functional class, tri-rheostat axis and per-disease frequency. Importance level: I model-prioritized (211 tri-rheostat) · II high-frequency (≥5%) · III standard.
dN/dS evolution analysis
dN/dS ratio (ω) across genes — ω > 1: possible positive selection (point estimate only), ω < 1: purifying selection, ω ≈ 1: neutral. Estimated by Nei–Gojobori with Jukes–Cantor correction; ω is unstable for short genes or low dS (few synonymous sites) and should be read together with gene length and synonymous/non-synonymous site counts.
Sample datasets
1,202 EBV whole genomes with metadata. Run accessions link to the SRA archive.
Download
Download aggregated data tables (CSV/TSV). Requires a registered, approved account.
Phylogenetic tree
Maximum-likelihood tree of 1,176 EBV whole genomes (leaves coloured by disease). Scroll to zoom, drag to pan.
Submit data
Have new EBV whole-genome sequencing data? Submit the sample metadata below — our team will review and integrate it into the database.
Data Updates & Variant Monitoring
Database status, new sample/position monitoring, and priority variant alerts.
New samples in database
Recently added EBV whole genomes (CRA038738 / PRJCA057887).
New typing positions
Positions newly reaching the ≥3-sample threshold after the latest re-filtering.
Priority variants for functional follow-up
Model-prioritized candidate positions in genes with possible positive selection (ω>1, point estimate only; potential functional / disease-relevant variants).
REST API
Interactive docs: Swagger UI · OpenAPI JSON. All endpoints are versioned under /api/v1/ (e.g. /api/v1/gene/BCRF1); the unversioned /api/ prefix remains as a backward-compatible alias.
| Endpoint | Description |
|---|---|
GET /api/gene/{name} | Gene annotation, evolution strategy, dN/dS |
GET /api/variant?gene=&axis=&limit= | Variants filtered by gene / tri-rheostat axis |
GET /api/disease | All 14 disease categories with tri-rheostat recipes |
GET /api/freq?variant_id=&position= | Per-disease mutation frequency of a position |
GET /api/shap?disease=&top_n= | Top tri-rheostat positions by |SHAP| for a disease |
GET /api/download/{table}?fmt=csv|tsv | Download any table |
GET /api/axis_radar?disease= | Tri-rheostat radar data |
GET /api/heatmap?n= | Top-position × disease frequency matrix |
GET /api/statistics | Disease & geographic distribution (donut charts) |
Example
curl "https://ebvatlas.com/api/v1/gene/BCRF1"Disease Comparison
Compare two diseases — three-axis recipes, shared/unique positions and per-position statistics (n/N, odds ratio, 95% CI, P value, FDR).
Cross-disease Comparison
Three-axis Encyclopedia
Disease Profiles
Model schematic
Three axes — immune evasion, latency signalling and lytic activity — summarize EBV genomic patterns associated with disease categories, shown here as three illustrative dials.
The tri-rheostat model
The three axes — immune evasion, latency signalling and lytic activity — summarize EBV genomic patterns associated with disease categories. They are derived from model attribution (SHAP) and serve as a descriptive, interpretable framework, extending the field's long-standing latency-centric view without implying that these axes independently determine disease.
Axis composition
211 tri-rheostat positions across 51 genes, grouped into three axes (+1 unassigned).
| Axis | Positions | Genes | Key genes |
|---|---|---|---|
| Immune evasion | 17 | 2 | BCRF1, BNLF2a |
| Latency signalling | 51 | 9 | LMP-1, EBNA-1, LMP-2A/B, EBNA-2/3 |
| Lytic activity | 142 | 39 | BHLF1, BPLF1, BALF2, LF3 |
| Other (unassigned) | 1 | 1 | BALF1 |
Disease recipes
Each disease shows a unique three-axis recipe (immune / latent / lytic %).
| Disease | Immune evasion | Latency signalling | Lytic activity |
|---|---|---|---|
| NPC | 30.4% | 33.4% | 36.2% |
| GC | 4.6% | 13.7% | 81.7% |
| AITL | 0.9% | 39.3% | 59.8% |
| Healthy | 7.8% | 41.4% | 50.8% |
Tri-rheostat radar
Model Card
| Research question | Can disease identity be recovered from EBV genome variation? (exploratory, not diagnostic) |
|---|---|
| Input data | 1,140 EBV genomes (11 classes); 8,306 typing positions → 2,436 features (881 individual + 55 projection + 1,500 interaction). |
| Model structure | XGBoost multi-class classifier. |
| Training / validation | Nested 5-fold cross-validation; macro F1 0.574, macro AUC 0.960. |
| Per-class performance | Highly separable: NPC (F1 0.83), BL (F1 0.80). Poorly separable: DLBCL (0.18), HL (0.33), ENKL (0.35) — these overlap with related lymphoid malignancies. |
| External validation | Leave-one-cohort-out: macro F1 0.210 (disease signal is real but weaker than the random-split estimate; lymphoid classes generalize better than epithelial NPC). Huashan 26-sample hold-out (lymphoid): ENKL top-3 100%. |
| What the model does | Summarizes EBV genomic patterns associated with disease categories, and nominates SHAP-selected disease-discriminating positions. |
| What the model does NOT do | Does not establish causality, does not prove the three axes independently determine disease, and is not a standalone diagnostic panel. |
| SHAP interpretation | SHAP values are model-attribution contributions, not direct measurements of biological importance. |
| Limitations | Sample bias (85% East Asian); causality not established; prospective cohorts required; the three functional categories are an annotation scheme, not independent regulatory axes. |
| Model version | (pipeline ). |
Exploratory disease-association classifier Research use only
Paste a list of mutated genomic positions (one per line, or comma/space-separated; optional ref>alt). The EBV-TRM model maps them to the 8,306 typing positions and predicts the Top-3 EBV-associated disease classes.
Gene
← Back to browseIndependent, shareable gene page. Stable URL: #/gene/BCRF1
Variant
← Back to variant browserIndependent, shareable variant page. Stable URL: #/variant/8873
Knowledge Graph
Genes ↔ tri-rheostat axes ↔ diseases. Drag / scroll to explore; hover to inspect.
Variant Function Knowledge Base
The 211 tri-rheostat model-prioritized candidate positions, annotated with gene function, evolutionary selection (dN/dS) and SHAP contribution — linking variant → function → disease.
How to Cite
[Authors]. EBVAtlas: a tri-rheostat database of 1,202 EBV whole genomes for disease stratification. [Journal], [Year]. DOI: [pending].
Please also cite the underlying methods paper and the EBV-TRM model.
BibTeX
@misc{EBVAtlas2026,
title = {EBVAtlas: a tri-rheostat database of 1,202 EBV whole genomes for disease stratification},
author = {Wang, Jianxin and others},
year = {2026},
doi = {[pending]},
url = {https://ebvatlas.com/}
}
RIS
TY - DATA T1 - EBVAtlas: a tri-rheostat database of 1,202 EBV whole genomes for disease stratification AU - Wang, Jianxin PY - 2026 DO - [pending] UR - https://ebvatlas.com/ ER -
Contact
Use the Feedback page, or email: jianxinwang1994@163.com
What's new
2026-10 · v0.1.0 release
- Added variant lookup, dN/dS evolution analysis, gene variation landscape, and tri-rheostat model page.
- Added user registration with download approval workflow.
- Initial release of EBV Genome Atlas (1,176 genomes · 8,232 typing positions · 211 tri-rheostat positions).
2026-10 · data update
- Added 26 new EBV whole genomes (CRA038738 / PRJCA057887, Huashan Hospital): 17 NK/T lymphoma, 8 CAEBV, 1 IM.
- Total genomes: 1,176 → 1,202. Typing positions: 8,232 → 8,306 (re-filtered at ≥3 samples).
2026-10 · new modules
- Added Data Updates & Variant Monitoring (new samples / new typing positions / priority variant alerts).
- Added Knowledge base (211 tri-rheostat positions with gene function, selection, SHAP).
- Added an online exploratory disease-association classifier (paste variant positions → Top-3 classification, with clinical-site restriction).
- Added world geographic-distribution map and variant importance grading (I / II / III).
Quick Tutorial
New here? Follow these steps to get started.
Use cases
Help & User Manual
EBV Genome Atlas (Data ) · Last pipeline run
1. About EBV Genome Atlas
EBV Genome Atlas is a curated, interactive database of Epstein–Barr virus (EBV) genomic variation for EBV-associated disease research. Built on EBV whole-genome sequencing data, it integrates a machine-learning disease-classification model with "tri-rheostat" functional annotations, helping researchers understand the virological basis of EBV disease diversity from whole-genome variation.
EBV infects more than 90% of adults worldwide, yet causes a strikingly diverse disease spectrum — from asymptomatic carriage to epithelial and lymphoid malignancies. EBV Genome Atlas introduces the tri-rheostat model: three axes — immune evasion, latency signalling and lytic activity — summarize EBV genomic patterns associated with disease categories.
2. Data Overview & Statistics
| Metric | Value | Note |
|---|---|---|
| EBV whole genomes | 1,202 | 1,176 public + 26 new samples |
| Typing positions | 8,306 | positions mutated in ≥3 samples |
| Tri-rheostat positions | 211 | immune evasion 17 · latency 51 · lytic 142 · unassigned 1 |
| Disease categories | 14 | incl. healthy carriers |
| Geographic regions | 9 | China / Japan / Africa / Europe / Southeast Asia / Oceania / Americas / South Korea / Unknown |
| Annotated genes | 87 | with UniProt annotation, sequences, functions |
| Tri-rheostat genes | 51 | genes involved in the tri-rheostat mechanism |
| SHAP positions | 895 | positions with SHAP contribution values |
3. Data Sources & Quality Control
Sources: SRA (NCBI), GenBank, DDBJ, GSA (China); new samples from CRA038738 / PRJCA057887 (Huashan Hospital, Fudan University).
Quality control: genome coverage ≥78% (≥75% for new data) and mean depth ≥100× (≥40× for new data).
Data scope: typing positions (8,306) are model features; tri-rheostat positions (211) are SHAP-prioritized candidate positions (not a diagnostic panel). Only aggregated frequencies are published — individual sequences are not distributed (privacy / ethics compliant).
4. Core Concepts & Methods
Tri-rheostat axes: immune evasion (BCRF1/vIL-10, BNLF2a) · latency signalling (LMP-1, EBNA family) · lytic activity (BHLF1, BPLF1, BALF2).
Machine-learning classification: 8,306 typing positions → 2,436 features (881 individual + 55 projection + 1,500 interaction) → XGBoost multi-class, Top-3 disease prediction with confidence; supports clinical-site-restricted prediction (epithelial / lymphoid).
dN/dS evolution: ω > 1 possible positive selection (point estimate only), ω < 1 purifying selection, ω ≈ 1 neutral.
5. Using the Interface
| Module | What you can do |
|---|---|
| Home | Overview, statistics, disease/geographic charts, release notes |
| Browse | Query by gene / tri-rheostat axis / disease / position; example queries |
| Variant | Frequency heatmap, variant lookup (with importance level), dN/dS evolution |
| Model | Tri-rheostat model, disease recipes, radar charts; exploratory disease-association classifier |
| Knowledge | Variant → function → disease knowledge base (211 model-prioritized candidate positions) |
| Datasets | Sample metadata (filter + pagination) |
| Monitoring | New samples / new typing positions / priority variant alerts |
| Download | Aggregated tables (CSV/TSV) — requires approved account |
| API | REST API with Swagger UI |
| Help | User manual + feedback |
6. REST API
Interactive docs: Swagger UI · OpenAPI JSON
curl "https://ebvatlas.com/api/gene/BCRF1"
curl "https://ebvatlas.com/api/statistics"7. FAQ
What is "tri-rheostat"? Three axes (immune evasion / latency / lytic) summarize EBV genomic patterns associated with disease categories.
Typing vs tri-rheostat positions? Typing positions (8,306) are model features; tri-rheostat positions (211) are SHAP-prioritized candidate positions.
Why register to download? Downloads require a registered, approved account for data governance.
Individual sequences? No — only aggregated frequencies are published.
Feedback
Questions or feedback? Leave a message — we will get back to you by email.
Website visits
Total page visits since launch.
User management
Review and approve registered accounts.
Data submissions
Review user-submitted EBV sample metadata.