"""
TopoGrammar - Grammar-Aware 3D Genome Engine Demo
==================================================
Hugging Face Spaces Demo for TopoGrammar v2.1.0
The Industry's First Grammar-Aware Engine for Balanced Structural Variants
Copyright (c) 2026 Bryan Daugherty, Gregory Ward & Shawn Ryan. All Rights Reserved.
This demo showcases TopoGrammar's capabilities without exposing proprietary algorithms.
All results are pre-computed demonstrations for educational purposes.
"""
import gradio as gr
import hashlib
import random
import time
import httpx
import json
import os
from datetime import datetime
from typing import Optional, Dict, List, Tuple
# =============================================================================
# BSV Blockchain Configuration
# =============================================================================
BSV_API_URL = "https://simplebsv.codenlighten.org"
BSV_API_KEY = os.getenv("BSV_API_KEY") # Set via HF Secrets
def publish_to_bsv(data: Dict) -> Optional[str]:
"""
Publish attestation data to BSV blockchain via SimpleBSV API.
Args:
data: Dictionary containing attestation data (no trade secrets)
Returns:
Transaction ID if successful, None otherwise
"""
# Guard clause: fail gracefully if no key configured
if not BSV_API_KEY:
print("BSV_API_KEY not configured - running in offline mode")
return None
try:
headers = {
"Content-Type": "application/json",
"x-api-key": BSV_API_KEY
}
# Use ?wait=true for synchronous response with txid
response = httpx.post(
f"{BSV_API_URL}/publish/json?wait=true",
headers=headers,
json={"json": data},
timeout=30.0
)
if response.status_code == 200:
result = response.json()
return result.get("txid")
else:
print(f"BSV API error: {response.status_code} - {response.text}")
return None
except Exception as e:
print(f"BSV publish error: {e}")
return None
# =============================================================================
# Demo Configuration - Pre-computed results (no trade secrets exposed)
# =============================================================================
# Demo VUS Examples with pre-computed classifications
VUS_EXAMPLES = {
"brca1_boundary": {
"name": "BRCA1 TAD Boundary Disruption",
"variant_id": "chr17:43,044,295 G>A",
"gene": "BRCA1",
"initial_class": "VUS",
"final_class": "Pathogenic",
"confidence": 0.93,
"mechanism": "TAD Boundary Disruption",
"evidence_codes": ["PS3_insulation", "PM1_boundary", "PS3_ctcf"],
"insulation_change": -0.52,
"description": "This intronic variant disrupts a CTCF binding site at a critical TAD boundary, causing enhancer-promoter miscommunication affecting BRCA1 expression.",
},
"myc_enhancer_hijack": {
"name": "MYC Enhancer Hijacking",
"variant_id": "chr8:128,750,000 inv(500kb)",
"gene": "MYC",
"initial_class": "VUS",
"final_class": "Pathogenic",
"confidence": 0.89,
"mechanism": "Enhancer Hijacking",
"evidence_codes": ["PS3_neoloop", "PS2_oncogene", "PM1_boundary"],
"insulation_change": -0.68,
"description": "Balanced inversion creates a neo-loop connecting MYC to a hijacked super-enhancer, causing oncogene activation without copy number change.",
},
"sonic_hedgehog": {
"name": "SHH Limb Enhancer",
"variant_id": "chr7:156,584,000 del(50kb)",
"gene": "SHH",
"initial_class": "VUS",
"final_class": "Likely Pathogenic",
"confidence": 0.85,
"mechanism": "Enhancer Deletion",
"evidence_codes": ["PM1_regulatory", "PP3_conservation", "PS3_insulation"],
"insulation_change": -0.41,
"description": "Deletion removes the ZRS limb enhancer from SHH regulatory domain, disrupting developmental gene expression.",
},
"benign_intronic": {
"name": "Benign Intronic SNP",
"variant_id": "chr12:25,398,284 C>T",
"gene": "KRAS",
"initial_class": "VUS",
"final_class": "Benign",
"confidence": 0.91,
"mechanism": "No 3D Impact",
"evidence_codes": ["BS1_frequency", "BP4_no_disruption"],
"insulation_change": 0.02,
"description": "Deep intronic variant with no effect on TAD structure, CTCF binding, or regulatory grammar. Common in population databases.",
},
"tp53_scramble": {
"name": "TP53 Grammar Scrambling",
"variant_id": "chr17:7,670,000 inv(120kb)",
"gene": "TP53",
"initial_class": "VUS",
"final_class": "Pathogenic",
"confidence": 0.96,
"mechanism": "Regulatory Grammar Scrambling",
"evidence_codes": ["PS3_semantic", "PS3_insulation", "PM1_boundary", "PP3_conservation"],
"insulation_change": -0.71,
"description": "Balanced inversion scrambles the regulatory grammar of TP53, inverting enhancer-promoter orientation and disrupting tumor suppressor expression.",
},
}
# Demo TAD examples
TAD_EXAMPLES = {
"chr21_dscr": {
"name": "Chromosome 21 - Down Syndrome Critical Region",
"region": "chr21:35,000,000-40,000,000",
"n_tads": 4,
"n_boundaries": 5,
"ctcf_sites": 12,
"genes": ["DSCR1", "DSCR3", "DSCR4", "RUNX1"],
"boundary_strength": [0.85, 0.92, 0.78, 0.88, 0.81],
},
"chr7_egfr": {
"name": "Chromosome 7 - EGFR Locus",
"region": "chr7:55,000,000-56,500,000",
"n_tads": 3,
"n_boundaries": 4,
"ctcf_sites": 8,
"genes": ["EGFR", "LANCL2", "VOPP1"],
"boundary_strength": [0.91, 0.87, 0.94, 0.82],
},
"chr8_myc": {
"name": "Chromosome 8 - MYC Oncogene",
"region": "chr8:127,500,000-129,500,000",
"n_tads": 2,
"n_boundaries": 3,
"ctcf_sites": 6,
"genes": ["MYC", "PVT1"],
"boundary_strength": [0.96, 0.89, 0.93],
},
}
# Demo Neo-Loop examples
NEOLOOP_EXAMPLES = {
"burkitt_myc": {
"name": "Burkitt Lymphoma - MYC Translocation",
"sv_type": "Translocation t(8;14)",
"oncogene": "MYC",
"hijacked_enhancer": "IGH Super-Enhancer",
"loop_strength": 0.92,
"clinical_priority": "Critical",
"cancer_type": "Burkitt Lymphoma",
"description": "Classic t(8;14) translocation juxtaposes MYC with immunoglobulin heavy chain enhancers, creating pathogenic neo-loop.",
},
"ewing_ewsr1": {
"name": "Ewing Sarcoma - EWSR1-FLI1",
"sv_type": "Translocation t(11;22)",
"oncogene": "EWSR1-FLI1 fusion",
"hijacked_enhancer": "GGAA microsatellite enhancers",
"loop_strength": 0.88,
"clinical_priority": "Critical",
"cancer_type": "Ewing Sarcoma",
"description": "Fusion protein creates neo-loops at GGAA microsatellites, aberrantly activating developmental genes.",
},
"aml_runx1": {
"name": "AML - RUNX1 Disruption",
"sv_type": "Inversion inv(16)",
"oncogene": "CBFB-MYH11 fusion",
"hijacked_enhancer": "Myeloid enhancer cluster",
"loop_strength": 0.85,
"clinical_priority": "High",
"cancer_type": "Acute Myeloid Leukemia",
"description": "Pericentric inversion disrupts normal RUNX1 regulation, creating aberrant chromatin loops.",
},
}
# Benchmark comparisons
BENCHMARK_DATA = {
"tad_detection": {
"TopoGrammar (CTCF-gated)": {"accuracy": 91, "f1": 0.91, "subtad": True},
"HiCCUPS": {"accuracy": 80, "f1": 0.76, "subtad": False},
"Arrowhead": {"accuracy": 78, "f1": 0.74, "subtad": False},
"TopDom": {"accuracy": 75, "f1": 0.71, "subtad": False},
},
"vus_resolution": {
"reclassification_rate": 68,
"pathogenic_accuracy": 93,
"mean_confidence": 87,
},
"insulation_density": {
"major_tads": 9.11,
"all_boundaries": 15.57,
"improvement": 70.9,
},
}
# =============================================================================
# Visualization Functions
# =============================================================================
def create_contact_map_ascii(region: str, n_tads: int) -> str:
"""Create ASCII representation of a Hi-C contact map with TADs."""
size = 40
map_lines = []
# Create diagonal pattern with TAD blocks
tad_size = size // n_tads
for i in range(size):
row = ""
for j in range(size):
if abs(i - j) <= 2:
row += "██" # Diagonal
elif (i // tad_size) == (j // tad_size):
# Within same TAD
distance = abs(i - j)
if distance < tad_size // 2:
row += "▓▓"
elif distance < tad_size:
row += "░░"
else:
row += " "
else:
row += " "
map_lines.append(row)
return "\n".join(map_lines)
def create_insulation_profile(boundary_strengths: List[float]) -> str:
"""Create ASCII insulation score profile."""
lines = []
lines.append("Insulation Score Profile:")
lines.append("=" * 60)
for i, strength in enumerate(boundary_strengths):
bar_len = int(strength * 40)
bar = "█" * bar_len + "░" * (40 - bar_len)
lines.append(f"Boundary {i+1}: [{bar}] {strength:.2f}")
lines.append("=" * 60)
return "\n".join(lines)
def create_grammar_diagram(mechanism: str) -> str:
"""Create regulatory grammar visualization."""
if mechanism == "Regulatory Grammar Scrambling":
return """
┌─────────────────────────────────────────────────────────────┐
│ REGULATORY GRAMMAR ANALYSIS │
├─────────────────────────────────────────────────────────────┤
│ │
│ REFERENCE GRAMMAR: │
│ ┌──────┐ ┌──────────┐ ┌──────────┐ ┌──────┐ │
│ │CTCF+ │ → │ ENHANCER │ → │ PROMOTER │ → │ GENE │ │
│ └──────┘ └──────────┘ └──────────┘ └──────┘ │
│ "The enhancer activates the gene" │
│ │
│ VARIANT GRAMMAR (SCRAMBLED): │
│ ┌──────┐ ┌──────────┐ ┌──────────┐ ┌──────┐ │
│ │ GENE │ ← │ PROMOTER │ ← │ ENHANCER │ ← │CTCF- │ │
│ └──────┘ └──────────┘ └──────────┘ └──────┘ │
│ "Gene the activates enhancer the" ⚠ SCRAMBLED │
│ │
│ SEMANTIC BREAK SCORE: 0.85 (CRITICAL) │
│ • Token Disruption: 30% weight │
│ • Order Inversion: 40% weight │
│ • Orientation Flip: 30% weight │
│ │
└─────────────────────────────────────────────────────────────┘
"""
elif mechanism == "Enhancer Hijacking":
return """
┌─────────────────────────────────────────────────────────────┐
│ ENHANCER HIJACKING ANALYSIS │
├─────────────────────────────────────────────────────────────┤
│ │
│ NORMAL TOPOLOGY: │
│ ┌─────────────────┐ ┌─────────────────┐ │
│ │ TAD A │ │ TAD B │ │
│ │ [Enhancer]──────│─X───│──────[Oncogene] │ │
│ │ ↓ │ │ │ │
│ │ [Target Gene] │ │ │ │
│ └─────────────────┘ └─────────────────┘ │
│ BOUNDARY BLOCKS CONTACT │
│ │
│ AFTER STRUCTURAL VARIANT: │
│ ┌──────────────────────────────────────────┐ │
│ │ FUSED TAD │ │
│ │ [Enhancer]═══════════════════[Oncogene] │ │
│ │ ↓ NEO-LOOP FORMED ↓ │ │
│ │ [Target Gene] ⚠ ONCOGENE ACTIVATED ⚠ │ │
│ └──────────────────────────────────────────┘ │
│ │
│ LOOP STRENGTH: 0.89 | PRIORITY: CRITICAL │
│ │
└─────────────────────────────────────────────────────────────┘
"""
elif mechanism == "TAD Boundary Disruption":
return """
┌─────────────────────────────────────────────────────────────┐
│ TAD BOUNDARY DISRUPTION ANALYSIS │
├─────────────────────────────────────────────────────────────┤
│ │
│ NORMAL INSULATION: │
│ ████████████████ │ ████████████████ │
│ TAD A │ TAD B │
│ CTCF│CTCF │
│ ▲▲▲▲▲│▲▲▲▲▲ │
│ Strong Boundary │
│ │
│ AFTER VARIANT (CTCF SITE DISRUPTED): │
│ ████████████████ ████████████████ │
│ TAD A ░░░░░░░ TAD B │
│ ↓↓↓↓↓ │
│ Insulation Leak │
│ │
│ INSULATION CHANGE: -52% │
│ BOUNDARY LOSS: MAJOR │
│ │
└─────────────────────────────────────────────────────────────┘
"""
else:
return """
┌─────────────────────────────────────────────────────────────┐
│ 3D GENOME ANALYSIS │
├─────────────────────────────────────────────────────────────┤
│ │
│ No significant 3D structural impact detected. │
│ │
│ • TAD boundaries: INTACT │
│ • Insulation scores: NORMAL │
│ • Regulatory grammar: PRESERVED │
│ • CTCF binding: UNAFFECTED │
│ │
└─────────────────────────────────────────────────────────────┘
"""
def create_confidence_bars(evidence_codes: List[str], confidence: float) -> str:
"""Create evidence code visualization."""
lines = []
lines.append("\n### Evidence Summary\n")
for code in evidence_codes:
# Determine strength from code prefix
if code.startswith("PS"):
strength = "Strong"
bar = "████████████████████"
color = "🟢"
elif code.startswith("PM"):
strength = "Moderate"
bar = "████████████░░░░░░░░"
color = "🟡"
elif code.startswith("PP"):
strength = "Supporting"
bar = "████████░░░░░░░░░░░░"
color = "🟠"
elif code.startswith("BS") or code.startswith("BP"):
strength = "Benign"
bar = "████████████████████"
color = "🔵"
else:
strength = "Unknown"
bar = "░░░░░░░░░░░░░░░░░░░░"
color = "⚪"
lines.append(f"{color} **{code}** ({strength}): `{bar}`")
lines.append(f"\n**Overall Confidence**: {confidence:.0%}")
return "\n".join(lines)
def generate_bsv_attestation(variant_id: str, classification: str, confidence: float) -> str:
"""Generate and publish BSV blockchain attestation."""
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S UTC")
# Create attestation data (no trade secrets - only results summary)
attestation_data = {
"platform": "TopoGrammar",
"version": "v2.1.0",
"timestamp": timestamp,
"analysis_type": "VUS_Resolution",
"variant_id": variant_id,
"classification": classification,
"confidence": round(confidence, 2),
"attestation_type": "demo"
}
# Create hash of the attestation data
data_str = json.dumps(attestation_data, sort_keys=True)
query_hash = hashlib.sha256(data_str.encode()).hexdigest()[:16]
full_hash = hashlib.sha256(data_str.encode()).hexdigest()
# Publish to BSV blockchain
txid = publish_to_bsv(attestation_data)
if txid:
# Real blockchain attestation
whatsonchain_url = f"https://whatsonchain.com/tx/{txid}"
status_line = f"Status: ✓ RECORDED ON BSV MAINNET"
txid_display = txid[:20] + "..." if len(txid) > 20 else txid
verify_section = f"""║ Transaction ID: {txid_display}
║ ║
║ 🔗 Verify on WhatsOnChain: ║
║ {whatsonchain_url[:54]}"""
else:
# Fallback if API fails
status_line = "Status: ⚠ OFFLINE MODE (BSV API unavailable)"
verify_section = f"""║ Data Hash: {full_hash[:32]}...
║ ║
║ ℹ Blockchain recording temporarily unavailable. ║
║ Result hash preserved for later attestation."""
return f"""
```
╔══════════════════════════════════════════════════════════════╗
║ TOPOGRAMMAR BSV ATTESTATION CERTIFICATE ║
╠══════════════════════════════════════════════════════════════╣
║ ║
║ Query Hash: {query_hash} ║
║ Timestamp: {timestamp} ║
║ Model Version: TopoGrammar v2.1.0 ║
║ ║
║ ───────────────────────────────────────────────────────── ║
║ ║
║ Variant: {variant_id}
║ Classification: {classification}
║ Confidence: {confidence:.1%}
║ ║
║ ───────────────────────────────────────────────────────── ║
║ ║
║ {status_line}
║ Network: BSV Mainnet ║
║ ║
{verify_section}
║ ║
║ This attestation is immutably recorded on BSV blockchain. ║
║ No proprietary algorithms or trade secrets are published. ║
║ ║
╚══════════════════════════════════════════════════════════════╝
```
"""
def create_share_links(variant_id: str, classification: str) -> str:
"""Create social sharing buttons."""
import urllib.parse
text = f"🧬 TopoGrammar reclassified {variant_id} as {classification}! Grammar-aware 3D genome analysis for precision medicine. #Genomics #AI #PrecisionMedicine"
twitter_url = f"https://twitter.com/intent/tweet?text={urllib.parse.quote(text)}&url=https://huggingface.co/spaces/GotThatData/TopoGrammar"
linkedin_url = f"https://www.linkedin.com/sharing/share-offsite/?url=https://huggingface.co/spaces/GotThatData/TopoGrammar"
return f"""
"""
# =============================================================================
# Main Demo Functions
# =============================================================================
def run_vus_analysis(example_key: str) -> str:
"""Run VUS resolution demo."""
if example_key not in VUS_EXAMPLES:
return "❌ Example not found. Please select a valid example."
# Simulate processing delay
time.sleep(1.5)
example = VUS_EXAMPLES[example_key]
# Build result markdown
result = f"""
# 🧬 VUS Resolution Analysis
## Variant Information
| Field | Value |
|-------|-------|
| **Variant ID** | `{example['variant_id']}` |
| **Gene** | {example['gene']} |
| **Initial Classification** | {example['initial_class']} |
---
## TopoGrammar Analysis Result
### Classification Update
| Before | → | After |
|--------|---|-------|
| **{example['initial_class']}** | 🔄 | **{example['final_class']}** |
### Confidence Score
{"█" * int(example['confidence'] * 20)}{"░" * (20 - int(example['confidence'] * 20))} **{example['confidence']:.0%}**
### Primary Mechanism
**{example['mechanism']}**
### 3D Structural Impact
- **Insulation Change**: {example['insulation_change']:+.0%}
---
## Mechanism Visualization
{create_grammar_diagram(example['mechanism'])}
---
{create_confidence_bars(example['evidence_codes'], example['confidence'])}
---
## Clinical Interpretation
{example['description']}
---
## BSV Verification
{generate_bsv_attestation(example['variant_id'], example['final_class'], example['confidence'])}
---
{create_share_links(example['variant_id'], example['final_class'])}
"""
return result
def run_tad_analysis(example_key: str) -> str:
"""Run TAD detection demo."""
if example_key not in TAD_EXAMPLES:
return "❌ Example not found. Please select a valid example."
time.sleep(1.0)
example = TAD_EXAMPLES[example_key]
result = f"""
# 🔬 TAD Detection Analysis
## Region Information
| Field | Value |
|-------|-------|
| **Region** | `{example['region']}` |
| **TADs Detected** | {example['n_tads']} |
| **Boundaries** | {example['n_boundaries']} |
| **CTCF Sites** | {example['ctcf_sites']} |
---
## Genes in Region
{', '.join([f"**{g}**" for g in example['genes']])}
---
## Contact Map Visualization
```
{create_contact_map_ascii(example['region'], example['n_tads'])}
```
---
## Boundary Strength Profile
```
{create_insulation_profile(example['boundary_strength'])}
```
---
## Detection Method
TopoGrammar uses **CTCF-gated boundary detection** which achieves:
- **91% TAD accuracy** (vs 80% for HiCCUPS)
- **0.91 F1 score** for boundary detection
- **Sub-TAD detection** capability
The CTCF-gating mechanism ensures boundaries are only called where:
1. Insulation score shows local minimum
2. CTCF binding evidence is present
3. Gradient analysis confirms boundary
---
## Benchmark Comparison
| Method | TAD Accuracy | Boundary F1 | Sub-TAD |
|--------|-------------|-------------|---------|
| **TopoGrammar** | **91%** | **0.91** | ✓ |
| HiCCUPS | 80% | 0.76 | ✗ |
| Arrowhead | 78% | 0.74 | ✗ |
| TopDom | 75% | 0.71 | ✗ |
"""
return result
def run_neoloop_analysis(example_key: str) -> str:
"""Run neo-loop detection demo."""
if example_key not in NEOLOOP_EXAMPLES:
return "❌ Example not found. Please select a valid example."
time.sleep(1.2)
example = NEOLOOP_EXAMPLES[example_key]
# Priority styling
if example['clinical_priority'] == "Critical":
priority_emoji = "🔴"
priority_style = "color: red; font-weight: bold;"
else:
priority_emoji = "🟡"
priority_style = "color: orange; font-weight: bold;"
result = f"""
# 🧪 Neo-Loop Detection Analysis
## Structural Variant
| Field | Value |
|-------|-------|
| **SV Type** | `{example['sv_type']}` |
| **Cancer Type** | {example['cancer_type']} |
| **Clinical Priority** | {priority_emoji} **{example['clinical_priority']}** |
---
## Oncogene Activation
### Activated Oncogene
**{example['oncogene']}**
### Hijacked Enhancer
**{example['hijacked_enhancer']}**
### Neo-Loop Strength
{"█" * int(example['loop_strength'] * 20)}{"░" * (20 - int(example['loop_strength'] * 20))} **{example['loop_strength']:.0%}**
---
## Mechanism Visualization
```
┌─────────────────────────────────────────────────────────────┐
│ NEO-LOOP FORMATION │
├─────────────────────────────────────────────────────────────┤
│ │
│ BEFORE: {example['sv_type']}
│ ┌────────────┐ ┌────────────┐ │
│ │ Enhancer │─ ─ X ─ ─│ Oncogene │ │
│ │ Domain A │ │ Domain B │ │
│ └────────────┘ └────────────┘ │
│ ↓ │
│ Normal Target │
│ │
│ AFTER: {example['sv_type']}
│ ┌──────────────────────────────────────┐ │
│ │ Enhancer ══════════ Oncogene │ │
│ │ ↓ NEO-LOOP ↓ │ │
│ │ ABERRANT ACTIVATION │ │
│ └──────────────────────────────────────┘ │
│ │
│ Loop Strength: {example['loop_strength']:.2f}
│ Priority: {example['clinical_priority']}
│ │
└─────────────────────────────────────────────────────────────┘
```
---
## Clinical Interpretation
{example['description']}
---
## Actionable Insights
Based on this neo-loop detection:
1. **Molecular Testing**: Confirm {example['sv_type']} by FISH or karyotyping
2. **Targeted Therapy**: Consider therapies targeting {example['oncogene']} pathway
3. **Clinical Trial**: Patient may be eligible for trials targeting this mechanism
4. **Monitoring**: Track {example['oncogene']} expression as biomarker
---
## BSV Verification
{generate_bsv_attestation(example['sv_type'], f"Neo-loop: {example['oncogene']}", example['loop_strength'])}
"""
return result
def show_benchmarks() -> str:
"""Show benchmark results."""
return f"""
# 📊 TopoGrammar Benchmarks
## TAD Detection Performance
| Method | TAD Accuracy | Boundary F1 | Sub-TAD Detection |
|--------|-------------|-------------|-------------------|
| **TopoGrammar (CTCF-gated)** | **91%** | **0.91** | ✓ Yes |
| HiCCUPS | 80% | 0.76 | ✗ No |
| Arrowhead | 78% | 0.74 | ✗ No |
| TopDom | 75% | 0.71 | ✗ No |
---
## VUS Resolution Performance
| Metric | Value |
|--------|-------|
| **Reclassification Rate** | 68% of VUS variants |
| **Pathogenic Accuracy** | 93% |
| **Mean Confidence Score** | 87% |
---
## Insulation Density Improvement
| Boundary Set | Insulation Ratio | P-value |
|--------------|------------------|---------|
| Major TADs only | 9.11x | <0.0001 |
| **All Boundaries** | **15.57x** | **<0.0001** |
| **Improvement** | **+70.9%** | — |
---
## What Makes TopoGrammar Different
### 1. Grammar-Aware Architecture
Traditional tools see DNA as a string of letters. TopoGrammar sees it as **sentences with grammar**:
```
Reference: [CTCF+] [Enhancer] [Promoter] [Gene]
"The enhancer activates the gene"
Inversion: [Gene] [Promoter] [Enhancer] [CTCF-]
"Gene the activates enhancer the" ← SCRAMBLED
```
### 2. Physics + Semantics Concordance
When both physics (insulation collapse) AND semantics (grammar scramble) agree:
| Evidence Type | Alone | Concordant |
|--------------|-------|------------|
| Physics | PM1 (Moderate) | — |
| Semantics | PM1 (Moderate) | — |
| **Both** | — | **PS3 (Strong)** ← UPGRADE |
### 3. CTCF-Gated Detection
Unlike other tools, TopoGrammar only calls boundaries where:
- ✓ Insulation score shows local minimum
- ✓ CTCF binding evidence is present
- ✓ Gradient analysis confirms boundary
This reduces false positives by **40%** compared to insulation-only methods.
---
## Clinical Value Pillars
| Pillar | Clinical Value | Technical Foundation |
|--------|---------------|---------------------|
| **Architectural Fidelity** | Eliminates VUS by proving physical boundary collapse | PINN Physics (15.57x insulation) |
| **Semantic Intelligence** | Detects "scrambled" instructions in balanced inversions | Regulatory Grammar Encoder |
| **Privacy-First Growth** | Global model evolution without data leakage | Async Federated Learning |
| **Clinician Clarity** | High-level medical prose instead of raw math | LLM Interpretation Layer |
"""
# =============================================================================
# Gradio Interface
# =============================================================================
HEADER_MD = """
# 🧬 TopoGrammar
## The Industry's First Grammar-Aware 3D Genome Engine
**Balanced Structural Variants (BSVs)** - inversions, translocations, complex rearrangements - appear "silent" to standard sequencers because they don't change gene dosage. But they **scramble the regulatory grammar** that controls gene expression.
**TopoGrammar solves this.** It's the first engine that understands chromatin as a *language* with grammar rules that can be broken.
---
| Capability | Performance |
|------------|-------------|
| 🎯 VUS Reclassification | 68% of variants |
| 📊 Pathogenic Accuracy | 93% confidence |
| 🔬 TAD Detection F1 | 0.91 (vs 0.76 HiCCUPS) |
| ⚡ Sub-TAD Detection | Yes (unique capability) |
---
"""
ABOUT_MD = """
# ℹ️ About TopoGrammar
## Overview
TopoGrammar is part of the **OmniPrime Enterprise Platform**, integrating:
- **TopoGrammar v2.1.0** - Grammar-Aware 3D Genome Engine
- **BioPrime v4.0 "Golden"** - Physics-First Molecular Docking
Together, they enable a seamless **Patient Genome → Drug Candidate** workflow.
---
## Core Innovation
### Regulatory Grammar Analysis
TopoGrammar treats regulatory elements as a **language**:
```
CTCF → Enhancer → Promoter → Gene
"The enhancer activates the gene"
```
When structural variants **scramble** this grammar, TopoGrammar detects it:
```
Gene ← Promoter ← Enhancer ← CTCF
"Gene the activates enhancer the" ⚠ PATHOGENIC
```
### Semantic Break Score
Quantifies regulatory disruption:
- **Token Disruption (30%)**: Elements removed or duplicated
- **Order Inversion (40%)**: Sequence rearranged
- **Orientation Flip (30%)**: Strand direction reversed
---
## Technology Stack
- **Physics-Informed Neural Networks (PINNs)** for 3D reconstruction
- **CTCF-Gated Boundary Detection** for precise TAD calling
- **Federated Learning** for privacy-preserving multi-site training
- **LLM Interpretation** for clinical reporting
- **BSV Blockchain** for result attestation
---
## Creators
- **Bryan Daugherty**
- **Gregory Ward**
- **Shawn Ryan**
---
## Learn More
🌐 [bioprime.one](https://bioprime.one) | 🧬 [OmniPrime Platform](https://github.com/Saifullah62/OmniPrime_v1.0)
---
**Copyright (c) 2026 Bryan Daugherty, Gregory Ward & Shawn Ryan. All Rights Reserved.**
*This demo showcases TopoGrammar capabilities. Actual clinical use requires the full OmniPrime Enterprise Platform.*
"""
# Custom CSS
CUSTOM_CSS = """
.gradio-container {
background: linear-gradient(135deg, #1a1a2e 0%, #16213e 50%, #0f3460 100%);
}
.gr-button-primary {
background: linear-gradient(90deg, #00d4ff, #00ff88) !important;
border: none !important;
}
.gr-button-secondary {
background: linear-gradient(90deg, #667eea, #764ba2) !important;
border: none !important;
color: white !important;
}
"""
# Build the interface
with gr.Blocks(
title="TopoGrammar - Grammar-Aware 3D Genome Engine",
theme=gr.themes.Base(
primary_hue="cyan",
secondary_hue="purple",
neutral_hue="slate",
),
css=CUSTOM_CSS,
) as demo:
gr.Markdown(HEADER_MD)
with gr.Tabs():
# Tab 1: VUS Resolution
with gr.TabItem("🧬 VUS Resolution"):
gr.Markdown("""
## Variant of Uncertain Significance → Clinical Classification
Select a demo variant to see how TopoGrammar reclassifies VUS using 3D genome analysis.
""")
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("### Select Example")
vus_brca1 = gr.Button("🔴 BRCA1 Boundary Disruption", variant="secondary")
vus_myc = gr.Button("🔴 MYC Enhancer Hijacking", variant="secondary")
vus_shh = gr.Button("🟡 SHH Limb Enhancer", variant="secondary")
vus_tp53 = gr.Button("🔴 TP53 Grammar Scrambling", variant="secondary")
vus_benign = gr.Button("🟢 Benign Intronic SNP", variant="secondary")
with gr.Column(scale=3):
vus_output = gr.Markdown("*Select an example to run VUS analysis*")
vus_brca1.click(fn=lambda: run_vus_analysis("brca1_boundary"), outputs=vus_output)
vus_myc.click(fn=lambda: run_vus_analysis("myc_enhancer_hijack"), outputs=vus_output)
vus_shh.click(fn=lambda: run_vus_analysis("sonic_hedgehog"), outputs=vus_output)
vus_tp53.click(fn=lambda: run_vus_analysis("tp53_scramble"), outputs=vus_output)
vus_benign.click(fn=lambda: run_vus_analysis("benign_intronic"), outputs=vus_output)
# Tab 2: TAD Detection
with gr.TabItem("🔬 TAD Detection"):
gr.Markdown("""
## Topologically Associating Domain Detection
See how TopoGrammar detects TAD boundaries with CTCF-gating for 91% accuracy.
""")
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("### Select Region")
tad_dscr = gr.Button("Chr21 - Down Syndrome Region", variant="secondary")
tad_egfr = gr.Button("Chr7 - EGFR Locus", variant="secondary")
tad_myc = gr.Button("Chr8 - MYC Oncogene", variant="secondary")
with gr.Column(scale=3):
tad_output = gr.Markdown("*Select a region to analyze TAD structure*")
tad_dscr.click(fn=lambda: run_tad_analysis("chr21_dscr"), outputs=tad_output)
tad_egfr.click(fn=lambda: run_tad_analysis("chr7_egfr"), outputs=tad_output)
tad_myc.click(fn=lambda: run_tad_analysis("chr8_myc"), outputs=tad_output)
# Tab 3: Neo-Loop Detection
with gr.TabItem("🧪 Neo-Loop Detection"):
gr.Markdown("""
## Cancer Neo-Loop & Enhancer Hijacking Detection
Identify oncogene activation through structural variant-induced neo-loops.
""")
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("### Select Cancer Example")
neo_burkitt = gr.Button("🔴 Burkitt Lymphoma (MYC)", variant="secondary")
neo_ewing = gr.Button("🔴 Ewing Sarcoma (EWSR1)", variant="secondary")
neo_aml = gr.Button("🟡 AML (RUNX1)", variant="secondary")
with gr.Column(scale=3):
neo_output = gr.Markdown("*Select a cancer example to detect neo-loops*")
neo_burkitt.click(fn=lambda: run_neoloop_analysis("burkitt_myc"), outputs=neo_output)
neo_ewing.click(fn=lambda: run_neoloop_analysis("ewing_ewsr1"), outputs=neo_output)
neo_aml.click(fn=lambda: run_neoloop_analysis("aml_runx1"), outputs=neo_output)
# Tab 4: Benchmarks
with gr.TabItem("📊 Benchmarks"):
gr.Markdown(show_benchmarks())
# Tab 5: About
with gr.TabItem("ℹ️ About"):
gr.Markdown(ABOUT_MD)
gr.Markdown("---")
gr.Markdown("""
**TopoGrammar v2.1.0** | Part of **OmniPrime Enterprise Platform**
[🌐 bioprime.one](https://bioprime.one) | [🧬 GitHub](https://github.com/Saifullah62/OmniPrime_v1.0) | [📧 Contact](mailto:info@bioprime.one)
*This is a demonstration. Clinical use requires the full OmniPrime Enterprise Platform.*
""")
if __name__ == "__main__":
demo.launch()