vivamed/Bio-Atlas

---
license: cc-by-sa-3.0
task_categories:
- text-classification
- question-answering
language:
- en
tags:
- biology
- drug-discovery
- genomics
- biomedical
- knowledge-graph
- sql
size_categories:
- 100M<n<1B
---

# 🧬 BioAtlas: The Most Comprehensive Biomedical Knowledge Base

**490+ million rows integrating genetics, expression, perturbations, drugs, and disease into one harmonized, queryable database.**

[](https://creativecommons.org/licenses/by-sa/3.0/)
[]()
[]()
[]()

---

## 📖 Introduction: Solving the Biomedical Data Integration Crisis

Modern biomedical research produces incredible datasets, but they're **isolated silos**:

- GWAS Catalog has genetics, but no gene expression
- LINCS has drug perturbations, but no genetic evidence
- ChEMBL has drugs, but no patient data or genomics
- Open Targets connects some pieces, but missing perturbation data

**The result?** Researchers spend months manually linking databases just to ask: *"Which drugs target genetically validated disease genes AND have good safety profiles?"*

### BioAtlas Solves This

**One PostgreSQL database. 40+ sources integrated. 490M+ rows. Unified IDs. Harmonized coordinates. Cross-platform normalized.**

Query across genetics + expression + perturbations + drugs + safety in **one SQL statement**. What takes months with raw databases takes seconds with BioAtlas.

---

## 🏆 Complete Coverage: Drugs, Diseases, Pathways, Mechanisms

**BioAtlas provides EVERYTHING about drugs, diseases, pathways, and mechanisms. Competitors give you fragments.**

### Drug Coverage (COMPLETE)

| What You Need | BioAtlas | Clarivate | STRING | Open Targets |
|---|---|---|---|---|
| **Drug Information** | ✅ 10,380 drugs | ⚠️ Limited | ❌ None | ⚠️ Limited |
| **Drug Identifiers** | ✅ 27,927 (ChEMBL, PubChem) | ⚠️ Some | ❌ None | ⚠️ Some |
| **Drug-Target** | ✅ 24,987 (11 sources merged) | ✅ Yes | ❌ None | ✅ Yes |
| **Binding Affinities** | ✅ 1,353,181 (pChEMBL unified) | ⚠️ Limited | ❌ None | ⚠️ Some |
| **Drug-Disease** | ✅ 47,642 indications | ✅ Yes | ❌ None | ✅ Yes |
| **Adverse Events** | ✅ 1,445,327 (FDA FAERS) | ✅ Yes | ❌ None | ⚠️ Limited |
| **Drug Perturbations** | ✅ 720K LINCS + 1.5M Tahoe | ⚠️ Limited | ❌ None | ❌ None |
| **Drug Screening** | ✅ 8.6M combo + 3.8M PRISM | ❌ None | ❌ None | ❌ None |
| **Activity Scores** | ✅ 204M TF/pathway | ❌ None | ❌ None | ❌ None |

**BioAtlas:** COMPLETE drug knowledge (mechanisms, targets, safety, perturbations, screening)

---

### Disease Coverage (COMPLETE)

| What You Need | BioAtlas | Clarivate | STRING | Open Targets |
|---|---|---|---|---|
| **Disease Ontology** | ✅ 30,259 (MONDO) | ✅ Yes | ❌ None | ✅ Yes |
| **Disease Definitions** | ✅ 22,205 detailed | ⚠️ Some | ❌ None | ⚠️ Some |
| **Disease Cross-Refs** | ✅ 104,423 (ICD, OMIM, etc) | ✅ Yes | ❌ None | ✅ Yes |
| **Gene-Disease** | ✅ 1,141,217 associations | ✅ Yes | ❌ None | ✅ Yes |
| **Disease-Phenotype** | ✅ 272,246 (HPO) | ⚠️ Limited | ❌ None | ✅ Yes |
| **Genetic Variants** | ✅ 299,791 curated | ⚠️ Limited | ❌ None | ✅ Yes |
| **GWAS Associations** | ✅ 112,823 variant-disease | ⚠️ Limited | ❌ None | ✅ Yes |
| **Colocalization** | ✅ 80M causal genes | ❌ None | ❌ None | ⚠️ Few |

**BioAtlas:** COMPLETE disease knowledge (genetics, phenotypes, variants, colocalization)

---

### Pathway & Mechanism Coverage (COMPLETE)

| What You Need | BioAtlas | Clarivate | STRING | Open Targets |
|---|---|---|---|---|
| **Pathway Definitions** | ✅ 2,781 (Reactome) | ✅ Yes | ❌ None | ✅ Yes |
| **Pathway Members** | ✅ 136,614 gene-pathway | ✅ Yes | ❌ None | ✅ Yes |
| **Pathway Footprints** | ✅ 252,769 (PROGENy) | ❌ None | ❌ None | ❌ None |
| **TF Regulation** | ✅ 31,953 edges (DoRothEA) | ⚠️ Limited | ❌ None | ⚠️ Limited |
| **Ligand-Receptor** | ✅ 20,911,544 pairs | ⚠️ Limited | ❌ None | ⚠️ Some |
| **Signaling** | ✅ 29,581 (OmniPath) | ✅ Yes | ⚠️ PPI | ⚠️ Limited |
| **GO Annotations** | ✅ 343,641 | ✅ Yes | ✅ Yes | ✅ Yes |
| **TF/Pathway Activities** | ✅ 204M scores | ❌ None | ❌ None | ❌ None |

**BioAtlas:** COMPLETE mechanism knowledge (pathways, TFs, L-R, signaling, activities)

---

### Integration & Normalization (UNIQUE)

| Feature | BioAtlas | Everyone Else |
|---|---|---|
| **All-in-One Database** | ✅ SQL joins across 40+ sources | ❌ Separate downloads |
| **ID Harmonization** | ✅ Ensembl↔ChEMBL↔MONDO | ⚠️ Manual mapping |
| **DMSO Normalization** | ✅ Multi-level cascade | ❌ Raw data |
| **pChEMBL Scale** | ✅ 1.35M unified | ⚠️ Mixed units |
| **GRCh38 Coordinates** | ✅ All harmonized | ⚠️ Mixed builds |
| **Cross-Platform Activities** | ✅ 204M comparable | ❌ Platform-specific |
| **Multi-Source Merges** | ✅ Adverse events, L-R, targets | ❌ Pick one source |
| **Colocalization** | ✅ 80M precomputed | ❌ Compute yourself |
| **Local SQL** | ✅ Full access | ⚠️ APIs/Web only |
| **Price** | **FREE** | $100K-$200K/year |

---

### The Bottom Line:

**For DRUGS:** BioAtlas has targets (25K), affinities (1.35M), indications (47K), safety (1.45M), perturbations (2.2M), screening (12M) - **ALL in one place**

**For DISEASES:** BioAtlas has ontology (30K), genetics (1.14M associations), phenotypes (272K), variants (299K), colocalization (80M) - **COMPLETE coverage**

**For PATHWAYS:** BioAtlas has definitions (2.7K), members (137K), footprints (253K), TF regulation (32K), activities (204M) - **FULL mechanism knowledge**

**For INTEGRATION:** BioAtlas is the ONLY one that connects all of these in one queryable database with advanced normalization

**Competitors:** Have 1-2 of these. BioAtlas has **ALL**.

---

## ⚡ Quick Start: Two Ways to Access

### Option A: 🐍 Python/Pandas (Quick Exploration)

**No database required!** Load individual tables directly with pandas:

```python
import pandas as pd

# Load any table directly from HuggingFace
drugs = pd.read_parquet("hf://datasets/vivamed/Bio-Atlas/parquet/drug.parquet")
diseases = pd.read_parquet("hf://datasets/vivamed/Bio-Atlas/parquet/disease.parquet")
drug_targets = pd.read_parquet("hf://datasets/vivamed/Bio-Atlas/parquet/drug_target.parquet")

# Load the flagship 80M colocalization data
coloc = pd.read_parquet("hf://datasets/vivamed/Bio-Atlas/parquet/coloc_bayesian.parquet")

print(f"Drugs: {len(drugs):,}") # 10,380
print(f"Diseases: {len(diseases):,}") # 30,259
print(f"Colocalization: {len(coloc):,}") # 79,926,756
```

**Available Parquet Files (35 tables, 116M+ rows):**

| Category | Tables | Key Files |
|----------|--------|-----------|
| **Core Entities** | 4 | `drug.parquet`, `disease.parquet`, `gene.parquet`, `pathway.parquet` |
| **Drug Data** | 7 | `drug_target.parquet`, `drug_adverse_event.parquet`, `drug_disease.parquet` |
| **Genetics** | 4 | `variant.parquet`, `variant_disease.parquet`, `coloc_bayesian.parquet` (80M!) |
| **Perturbations** | 5 | `l1000_signature.parquet`, `tahoe_activity.parquet`, `prism_response.parquet` |
| **Networks** | 5 | `lr_interaction.parquet` (21M), `tf_regulation.parquet`, `pathway_member.parquet` |
| **Ontologies** | 5 | `go_term.parquet`, `hpo_term.parquet`, `disease_hpo.parquet` |
| **Screening** | 3 | `drugcomb_summary.parquet`, `depmap_gene_dependency.parquet` |

---

### Option B: 🐘 PostgreSQL (Full Power)

**For complex multi-table queries and the complete 490M+ row database.**

#### Prerequisites
- PostgreSQL 14+ installed
- ~30 GB free disk space
- ~8 GB RAM minimum

#### 1. Download Files (Total ~26 GB)

```bash
# Core Knowledge Graph (14.2 GB) - Drugs, diseases, pathways, networks
huggingface-cli download vivamed/Bio-Atlas bioatlas_public_v1.0.dump

# LINCS Activity Scores (5.1 GB) - 202M perturbation scores
huggingface-cli download vivamed/Bio-Atlas bio_kg_v1.0.dump

# Colocalization Data (6.5 GB) - 80M GWAS colocalization tests
huggingface-cli download vivamed/Bio-Atlas coloc_bayesian.dump
```

#### 2. Load Database

```bash
# Create database
createdb bioatlas

# 1. Load Core Tables (SQL format) ~15 mins
psql -d bioatlas -f bioatlas_public_v1.0.dump

# 2. Load LINCS Activities (Binary format) ~10 mins
pg_restore -d bioatlas bio_kg_v1.0.dump

# 3. Load Colocalization (Binary format) ~10 mins
pg_restore -d bioatlas coloc_bayesian.dump
```

#### 3. Verify Installation

```sql
psql -d bioatlas -c "\dt" -- List tables (should see 79 tables)

psql -d bioatlas -c "SELECT COUNT(*) FROM drug;" -- Should be 10,380
psql -d bioatlas -c "SELECT COUNT(*) FROM l1000_activity;" -- Should be 202,282,258
psql -d bioatlas -c "SELECT COUNT(*) FROM coloc_bayesian;" -- Should be 79,926,756
```

---

## 🌍 Scope: The Breadth of Integration

### What We Integrated (40+ Databases)

**Genetics (2.5+ Billion Variant Associations):**
- GWAS Catalog: 443,634 trait studies
- UK Biobank: Full summary statistics
- **80 million colocalization tests** linking variants to causal genes
- GTEx: 415K tissue-specific eQTLs
- eQTL Catalogue: 90K multi-tissue eQTLs
- Open Targets: 1.14M gene-disease associations with L2G scores

**Gene Expression (85+ Million Measurements):**
- **LINCS L1000:** 57.6M gene measurements (720K signatures, 33,609 compounds)
- **CELLxGENE:** 28.2M single-cell measurements (714 cell types)
- **GTEx:** 2M tissue expression measurements
- **Tahoe-100M:** 1.5M activity scores from 100M cells (379 drugs, 50 cell lines)

**Drug Perturbations (204+ Million Activity Scores):**
- **LINCS L1000:** 202M TF/pathway activity scores
- **Tahoe-100M:** 1.55M activity scores (DMSO-normalized where available)
- **SC CRISPR:** 1.6M activities from 208K gene knockouts (4 studies)
- **Cross-platform harmonized** using same TF/pathway networks

**Drug Discovery (25K+ Interactions):**
- **ChEMBL:** 10K drugs, mechanisms, indications
- **BindingDB:** 1.35M binding affinities → pChEMBL unified scale
- **Drug-Target:** 25K interactions from 11 sources merged
- **Adverse Events:** 1.45M records (FDA FAERS)
- **Clinical Indications:** 47K drug-disease pairs

**Drug Screening (12+ Million Experiments):**
- **DrugComb:** 8.6M combination experiments (RAW data)
- **PRISM:** 3.8M drug sensitivity measurements
- **DepMap:** 1.3M CRISPR gene dependencies (essentiality)

**Interactions (21+ Million Edges):**
- **Ligand-Receptor:** 20.9M pairs (CellPhoneDB + CellChatDB)
- **Protein Signaling:** 29K interactions (OmniPath, commercial-filtered)
- **TF Regulation:** 32K edges (DoRothEA, commercial-filtered)
- **Pathway Membership:** 137K gene-pathway links

**Ontologies (Complete):**
- MONDO (30K diseases), HPO (19K phenotypes)
- GO (48K terms, 344K annotations)
- Cell Ontology (714 types), UBERON (78 tissues)
- 5,249 EFO-MONDO cross-references

---

## 📏 Scale: The Numbers

### Total Data Volume
- **490+ million documented rows**
- **79 core tables** (public release)
- **40+ data sources** harmonized
- **~26 GB compressed** (14.2 + 5.1 + 6.5 GB)

### Integration Achievements
- **43,608 genes** harmonized across Ensembl/HGNC/UniProt/Entrez
- **30,259 diseases** unified via MONDO ontology
- **10,380 drugs** linked across ChEMBL/Broad/PubChem
- **1.35M potencies** normalized to pChEMBL scale
- **All coordinates** standardized to GRCh38

### Unique at Scale
- **79,926,756 colocalization tests** (GWAS × eQTL Bayesian analysis)
- **63,332,122 with strong evidence** (H4 > 0.8) - identifies causal genes
- **204.5M activity scores** across 3 platforms (comparable framework)
- **1.45M adverse events** from FDA FAERS
- **20.9M ligand-receptor pairs** from 2 databases merged

---

## 🔧 Normalization: The Cleaning We Did

### 1. DMSO Cascade Normalization (Tahoe-100M)

**Problem:** Plate effects, batch effects, vehicle toxicity confound drug signals

**Our Multi-Level Solution:**

```
Cascade Matching Strategy:

Priority 1: Exact Match (~60% of scores)
- Same plate + cell line + time + feature
- drug_score - dmso_score = perfect batch correction
- Highest confidence scores

Priority 2: Plate Match (Additional contexts)
- Same plate + feature (different cell/time)
- drug_score - mean(plate_dmso)
- Corrects major plate effects

Priority 3: Raw Score (~40% of scores)
- No matching DMSO available
- Raw score with NO_CONTROL flag
- Still biologically meaningful, documented limitation
```

**Quality Flags:** Every score tagged with control availability
**User Choice:** Filter to normalized scores for highest confidence, or use all with documented caveats

### 2. pChEMBL Potency Unification

**Problem:** Binding data chaos across databases

**Before:**
- Ki in nM from ChEMBL
- Kd in μM from BindingDB
- IC50 in various units
- EC50 in mg/mL
- **Incomparable!**

**After:**

```
All → pChEMBL = -log10(molar)

Measurement Quality Ranking:
Ki (equilibrium) > Kd (dissociation) > IC50 (functional) > EC50 (functional)

Result: 1,353,181 directly comparable potencies
```

### 3. GWAS Harmonization to GRCh38

**Problem:** Different builds, coordinates, alleles across 443K studies

**Our Pipeline:**

```
1. Coordinate liftover to GRCh38
2. Variant ID standardization: chr:pos:ref:alt
3. Allele harmonization (consistent effect alleles)
4. QC filters:
- MAF > 0.01
- INFO > 0.8
- p < 5×10⁻⁸
5. Ready for colocalization
```

**Result:** Clean, queryable, colocalizable genetic data

### 4. Cross-Platform Activity Normalization

**Problem:** LINCS (978 genes), Tahoe (62K genes), CRISPR (variable) - how to compare?

**Solution:** Apply same networks (DoRothEA + PROGENy) to all platforms

| Platform | Input Genes | TF Activities | Pathway Activities | Comparability |
|---|---|---|---|---|
| LINCS L1000 | 978 | 202M scores | 10M scores | ✅ Same TFs/pathways |
| Tahoe-100M | 62,710 | 1.2M scores | 350K scores | ✅ Same TFs/pathways |
| SC CRISPR | 8-23K | 1.6M scores | - | ✅ Same TFs/pathways |

**Result:** TP53 activity in LINCS comparable to TP53 in Tahoe - cross-platform validation!

---

## 💡 Core Innovation: 99.6% Dimensional Reduction

**After integration, we solve the interpretation problem:**

### The Gene Expression Challenge
- 60,000+ genes measured per experiment
- 99% unchanged or noise
- Biologically overwhelming

### Our Network-Based Solution

```
Gene Expression (60,710 dimensions) - Noisy, uninterpretable
↓
Filter to network genes (~6,000 relevant)
↓
Apply DoRothEA (31,953 TF-target edges, 242 TFs)
↓
TF Activities (242 dimensions) - Regulatory interpretation
↓
Apply PROGENy (252,769 weights, 14 pathways)
↓
Pathway Activities (14 dimensions) - Mechanism interpretation
```

**Reduction:** 60,710 → 256 features (99.6%)
**Method:** ULM (Univariate Linear Model)
**Formula:** `activity = Σ(expression × sign) / √n_targets`

**Example - Paclitaxel:**
- **Before:** 60,710 gene changes (noise!)
- **After:** Trail pathway +2.8 (apoptosis), TP53 +1.9 (tumor suppressor)
- **Insight:** Mechanism-of-action clear!

---

## 🔥 What Only BioAtlas Has

### 1. 80 Million Colocalization Tests

**GWAS identifies variant-disease associations, but which gene is causal?**

**BioAtlas answer:** 79,926,756 Bayesian colocalization tests (GWAS × eQTL)

- 63,332,122 with strong causal evidence (H4 > 0.8)
- 70% have very strong evidence (H4 > 0.9)
- Precomputed (weeks of analysis)
- Instant causal gene queries

**Example Query:**

```sql
-- Find genes causally linked to Type 2 Diabetes
SELECT DISTINCT
c."rightStudyLocusId" as gene_study,
c.h4 as probability_causal,
c.chromosome
FROM coloc_bayesian c
WHERE c."leftStudyLocusId" LIKE '%diabetes%'
AND c.h4 > 0.9
ORDER BY c.h4 DESC
LIMIT 10;
```

### 2. Multi-Source Merges (Not Separate Downloads)

**Adverse Events:** FDA_FAERS = 1.45M (comprehensive)
**Ligand-Receptor:** CellPhoneDB + CellChatDB = 20.9M (comprehensive)
**Drug-Target:** ChEMBL + 10 sources = 25K (evidence-weighted)

### 3. Cross-Platform Harmonization

**Same TF/pathway framework across all perturbation datasets:**
- Compare LINCS drug → Tahoe drug → CRISPR gene knockout
- Validate mechanisms across platforms
- 204.5M scores using consistent networks

### 4. Complete Local SQL Access

**No APIs, no rate limits, no internet required**
- Download once, query forever
- Full PostgreSQL power
- Multi-hop joins across all 40+ sources

---

## 💻 Usage Examples

### Example 1: Find Drugs Targeting Genetically Validated Genes

```sql
-- Find FDA-approved drugs targeting genes with strong
-- genetic evidence for Alzheimer's disease
SELECT DISTINCT
d.drug_name,
d.max_phase,
g.hgnc_symbol,
dg.association_score,
dt.action_type
FROM drug d
JOIN drug_target dt ON d.molecule_chembl_id = dt.molecule_chembl_id
JOIN gene g ON dt.ensembl_gene_id = g.ensembl_gene_id
JOIN disease_gene dg ON g.ensembl_gene_id = dg.ensembl_gene_id
JOIN disease dis ON dg.mondo_id = dis.mondo_id
WHERE dis.disease_label ILIKE '%Alzheimer%'
AND d.fda_approved_us = TRUE
AND dg.association_score > 0.5
ORDER BY dg.association_score DESC;
```

### Example 2: Drug Mechanism-of-Action Analysis

```sql
-- What pathways does Paclitaxel activate across cell lines?
SELECT
tc.cell_line_name,
ta.feature_id as pathway,
AVG(ta.score) as mean_activity,
COUNT(*) as n_measurements
FROM tahoe_context tc
JOIN tahoe_activity ta ON tc.context_id = ta.context_id
WHERE tc.drug_name = 'Paclitaxel'
AND ta.provider = 'PROGENy'
AND ta.feature_type = 'pathway'
GROUP BY tc.cell_line_name, ta.feature_id
HAVING ABS(AVG(ta.score)) > 2.0
ORDER BY ABS(AVG(ta.score)) DESC;
```

### Example 3: Colocalization-Based Drug Repurposing

```sql
-- Find drugs targeting genes with strong colocalization
-- evidence for inflammatory bowel disease
SELECT DISTINCT
d.drug_name,
g.hgnc_symbol,
c.h4 as colocalization_prob,
dt.action_type
FROM coloc_bayesian c
JOIN gene g ON c."rightStudyLocusId" LIKE '%' || g.ensembl_gene_id || '%'
JOIN drug_target dt ON g.ensembl_gene_id = dt.ensembl_gene_id
JOIN drug d ON dt.molecule_chembl_id = d.molecule_chembl_id
WHERE c."leftStudyLocusId" LIKE '%inflammatory_bowel%'
AND c.h4 > 0.9
AND d.max_phase >= 2
LIMIT 20;
```

### Example 4: Cross-Platform Validation

```sql
-- Compare drug effects in LINCS vs Tahoe for the same compound
SELECT
'LINCS' as platform,
l.pert_iname as drug_name,
COUNT(*) as signatures
FROM l1000_signature l
WHERE l.molecule_chembl_id = 'CHEMBL1201259' -- Imatinib
GROUP BY l.pert_iname

UNION ALL

SELECT
'Tahoe',
tc.drug_name,
COUNT(DISTINCT tc.context_id)
FROM tahoe_context tc
WHERE tc.molecule_chembl_id = 'CHEMBL1201259'
GROUP BY tc.drug_name;
```

---

## 🔧 Advanced Features

### 1. Multi-Hop Graph Queries

BioAtlas enables complex multi-hop queries that would require downloading and linking dozens of separate databases:

```sql
-- Find drugs that:
-- 1. Target genes with colocalization evidence
-- 2. Have favorable safety profiles
-- 3. Are in clinical trials
SELECT
d.drug_name,
d.max_phase,
g.hgnc_symbol,
c.h4 as genetic_evidence,
COUNT(DISTINCT dae.meddra_pt) as adverse_event_count
FROM drug d
JOIN drug_target dt ON d.molecule_chembl_id = dt.molecule_chembl_id
JOIN gene g ON dt.ensembl_gene_id = g.ensembl_gene_id
JOIN coloc_bayesian c ON c."rightStudyLocusId" LIKE '%' || g.ensembl_gene_id || '%'
LEFT JOIN drug_adverse_event dae ON d.molecule_chembl_id = dae.molecule_chembl_id
WHERE c.h4 > 0.8
AND d.max_phase >= 2
GROUP BY d.drug_name, d.max_phase, g.hgnc_symbol, c.h4
HAVING COUNT(DISTINCT dae.meddra_pt) < 10 -- Good safety profile
ORDER BY c.h4 DESC
LIMIT 20;
```

### 2. Pathway Enrichment Analysis

```sql
-- Which pathways are enriched in genes associated with Type 2 Diabetes?
SELECT
p.pathway_label,
COUNT(DISTINCT pm.ensembl_gene_id) as n_genes,
AVG(dg.association_score) as avg_association
FROM disease dis
JOIN disease_gene dg ON dis.mondo_id = dg.mondo_id
JOIN pathway_member pm ON dg.ensembl_gene_id = pm.ensembl_gene_id
JOIN pathway p ON pm.pathway_id = p.pathway_id
WHERE dis.disease_label ILIKE '%diabetes%type 2%'
GROUP BY p.pathway_label
HAVING COUNT(DISTINCT pm.ensembl_gene_id) >= 5
ORDER BY COUNT(DISTINCT pm.ensembl_gene_id) DESC;
```

### 3. Drug Combination Synergy Analysis

```sql
-- Find synergistic drug combinations from DrugComb
SELECT
dc.drug_row,
dc.drug_col,
dc.cell_line_name,
dc.mean_synergy_zip,
dc.mean_synergy_bliss
FROM drugcomb_experiment dc
WHERE dc.mean_synergy_zip > 10 -- Strong synergy
AND dc.num_dose_points >= 16 -- Well-powered experiment
ORDER BY dc.mean_synergy_zip DESC
LIMIT 20;
```

---

## 📊 Data Dictionary

See [DATA_DICTIONARY.md](DATA_DICTIONARY.md) for complete table schemas and column descriptions.

**Quick Reference:**

| Category | Tables | Key Tables |
|---|---|---|
| **Core Entities** | 6 | drug, disease, gene, pathway |
| **Drug Data** | 9 | drug_target, drug_adverse_event, drug_disease |
| **Disease Data** | 3 | disease_gene, disease_hpo |
| **Colocalization** | 2 | coloc_bayesian, coloc_ecaviar |
| **LINCS L1000** | 2 | l1000_signature, l1000_activity |
| **Tahoe (scRNA)** | 3 | tahoe_activity, tahoe_context |
| **SC CRISPR** | 4 | sc_activity, sc_perturbation |
| **Drug Screening** | 9 | drugcomb_full_response, prism_response |
| **Networks** | 7 | lr_interaction, pathway_member, tf_regulation |
| **Ontologies** | 11 | hpo_term, go_term |
| **Variants/eQTLs** | 7 | variant, variant_disease, gtex_v10_eqtl_leads |
| **Gene Data** | 5 | gene_celltype_expression |

**Total:** 79 tables, 490M+ rows

---

## 🔥 What Only BioAtlas Has

### 1. 80 Million Colocalization Tests

**GWAS identifies regions, but colocalization identifies genes.**

- **79,926,756** Bayesian tests (GWAS × eQTL/pQTL/sQTL)
- **63,332,122** with strong causal evidence (H4 > 0.8)
- **70%** have very strong evidence (H4 > 0.9)
- Covers all 23 chromosomes
- Links 443K GWAS studies × 1.3M molecular QTL studies

**Why this matters:** A GWAS hit tells you "this region is associated with disease." Colocalization tells you "this specific gene is probably causal." That's the difference between a lead and a validated target.

### 2. Dimensionality Reduction (99.6%)

Gene expression is noisy (60,000 genes, 99% unchanged). We reduce it to **pathway activities**:

- **DoRothEA:** 242 Transcription Factors
- **PROGENy:** 14 Signaling Pathways

**Result:** Interpretable mechanism-of-action profiles for every drug.

**Example:**
```
Raw data: Gene X +1.2, Gene Y -0.8, Gene Z +0.3, ... (60,000 genes)
Reduced: MAPK pathway +3.2, TP53 activity +2.1, Trail pathway +1.8
```

### 3. Cross-Platform Activity Harmonization

**Same TF/pathway scores across all perturbation platforms:**

- **LINCS L1000:** 202M activity scores
- **Tahoe-100M:** 1.55M activity scores
- **SC CRISPR:** 1.6M activity scores

**Impact:** You can directly compare a drug's MAPK activation in LINCS to a gene knockout's MAPK effect in CRISPR. Cross-platform validation increases confidence 10x.

### 4. Multi-Source Integration

**Competitors force you to pick ONE source. We merged ALL sources:**

| Feature | Sources Merged | Result |
|---|---|---|
| Drug-Target | 11 sources | 24,987 interactions (evidence-weighted) |
| Adverse Events | FDA FAERS | 1.45M events (comprehensive) |
| Ligand-Receptor | CellPhoneDB + CellChatDB | 20.9M pairs (complete) |

---

## 📜 License & Attribution

**License:** [CC BY-SA 3.0](https://creativecommons.org/licenses/by-sa/3.0/)

### What You Can Do:
✅ Use for commercial drug discovery
✅ Build AI/ML models
✅ Create products and sell them
✅ Modify and redistribute the database

### What You Must Do:
⚠️ Provide attribution to BioAtlas
⚠️ Share derivative databases under CC BY-SA 3.0
⚠️ Include license notice

### Key Data Sources:
- **Open Targets Genetics:** Colocalization data (Apache 2.0)
- **ChEMBL:** Drug data (CC BY-SA 3.0)
- **LINCS L1000:** Perturbation data (CC BY 4.0)
- **FDA FAERS:** Adverse events (Public Domain)
- **Reactome/GO/MONDO:** Ontologies (CC BY 4.0)
- **DepMap, PRISM, DrugComb:** Screening data (CC BY 4.0)
- **GTEx, eQTL Catalogue:** eQTL data (Open Access)

*See [LICENSES_AND_ATTRIBUTIONS.md](LICENSES_AND_ATTRIBUTIONS.md) for complete details.*

---

## 📖 Citation

If you use BioAtlas in your research, please cite:

```bibtex
@dataset{bioatlas_2025,
title={BioAtlas: Integrated Multi-Omic Biomedical Knowledge Base},
author={Harris, Nicholas and Vivamed and EveryCure},
year={2025},
publisher={HuggingFace},
url={https://huggingface.co/datasets/vivamed/Bio-Atlas}
}
```

---

## 📚 Additional Documentation

- **[DATA_DICTIONARY.md](DATA_DICTIONARY.md)** - Complete table schemas and column descriptions
- **[RESEARCH_ARTICLE.md](RESEARCH_ARTICLE.md)** - Technical methodology and processing details
- **[schema.json](schema.json)** - Machine-readable schema for programmatic access
- **[LICENSES_AND_ATTRIBUTIONS.md](LICENSES_AND_ATTRIBUTIONS.md)** - Complete attribution for all 40+ sources

---

## 🤝 Support & Contact

**Issues?** Open an issue on this repository
**Questions?** Contact: [Your Contact Info]
**Commercial Use?** Fully allowed under CC BY-SA 3.0

---

## 🎉 Acknowledgments

BioAtlas integrates data from 40+ publicly available databases. We thank all the researchers, institutions, and consortia who make their data openly available:

- EMBL-EBI (ChEMBL, Open Targets)
- Broad Institute (LINCS, DepMap, PRISM)
- NHGRI-EBI (GWAS Catalog)
- GTEx Consortium
- Gene Ontology Consortium
- Monarch Initiative (MONDO, HPO)
- And many more...

**Your work makes open science possible.** 🙏

---

**BioAtlas: The most comprehensive biomedical knowledge base, freely available to accelerate drug discovery and precision medicine.** 🧬