🧬 GenomicOracle — AI × Quantum Genomic Analysis Platform

Precision oncology powered by Quantum Machine Learning, Multi-LLM Clinical Reasoning, and SOPHiA DDM™ Integration

📋 Overview

GenomicOracle is a full-stack genomic analysis platform built for the SOPHiA Genetics × ETH Zurich Future of Health Hackathon (May 8–10, 2026). It combines three pillars of modern precision oncology:

⚛️ Quantum Machine Learning — Variational Quantum Circuits (VQC) for pathogenicity scoring and Grover-inspired combinatorial search for multi-variant interactions (PennyLane)
🤖 Multi-LLM Clinical Reasoning — Evidence-based treatment recommendations and interactive clinical Q&A powered by Claude, Gemini, or GPT-4o
🔬 SOPHiA DDM™ Integration — Native VQS API connectivity and CSV export support for real-world genomic data from clinical labs

The platform ships with two complementary interfaces:

Interface	Port	Purpose
GenomicOracle (`main.py`)	`5000`	Full patient analysis — VCF upload, QML scoring, LLM reports, PDF export, AI chat
Genomic Co-Pilot (`GenomicSophia/app.py`)	`5001`	Clinical triage dashboard — bulk case ranking, 4-metric scoring, PubMed literature search

✨ Key Features

GenomicOracle — Full Patient Analysis

Feature	Description
🧬 VCF File Upload	Parse standard `.vcf` files or enter genomic variants manually
⚛️ Quantum ML Scoring	Variational Quantum Circuit (VQC) pathogenicity scoring via PennyLane
🔍 Grover-Inspired Search	Quantum combinatorial search across multi-variant state spaces for pathogenic interactions
🤖 LLM Treatment Reports	NCCN/ESMO-aligned clinical oncology reports generated by Claude, Gemini, or GPT-4o
💬 Interactive AI Chat	Context-aware clinical Q&A chatbot on the results page
🎓 VQC Training Dashboard	Live training visualization with loss curves and terminal logs
📄 PDF Report Export	Download formatted clinical genomics reports as PDF (xhtml2pdf)
🔒 Offline Fallback	Full expert rule-base when no API key is configured — covers EGFR, KRAS, BRAF, PIK3CA, ALK, BRCA, and more

Genomic Co-Pilot — Clinical Triage Dashboard

Feature	Description
📊 4-Metric Case Scoring	ABCD Prediction (35%) + ClinVar Evidence (25%) + Community Frequency (20%) + QA Confidence (20%)
🏥 Case Priority Ranking	🔴 Critical → 🟠 High → 🟡 Moderate → 🟢 Low urgency tiers
📁 Bulk CSV Analysis	Drop SOPHiA DDM™ exports into `/samples` — auto-loaded and ranked on startup
🔌 VQS API Integration	Live connection to SOPHiA DDM™ platform via dataset keys
🧠 AI Clinical Summaries	Optional Claude/Gemini/GPT-4o 2-sentence case summaries
⚡ Deterministic Summaries	Always-available rule-based 2-sentence summaries (no API key needed)
📚 PubMed Literature Search	Per-variant paper search via NCBI E-utilities with clickable paper cards
💊 Therapy Matching	21-gene actionable database with FDA-approved targeted therapies
🌓 Dark/Light Mode	Toggle between dark and light themes

CLI — Batch Variant Analysis

Feature	Description
📋 `variant_analysis_ai.py`	Batch-process up to 10 SOPHiA DDM™ CSV files via Claude API
🎯 AI Importance Scoring	0–100 importance score per sample with 2-sentence clinical summary
📊 CSV Output	Saves aggregated results to `ai_analysis.csv`

🏗️ Architecture

┌─────────────────────────────────────────────────────────────────────────┐
│                    GenomicOracle Platform                               │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                         │
│  ┌────────────────────────────────────────────────────────────────────┐  │
│  │  main.py — Full Patient Analysis (port 5000)                      │  │
│  │  ┌──────────────┬─────────────────┬──────────────────────────┐    │  │
│  │  │  VCF Parser  │  QML Module     │    LLM Module            │    │  │
│  │  │  (vcfpy)     │ (qml_module.py) │   (llm_module.py)        │    │  │
│  │  │              │                 │                          │    │  │
│  │  │              │ • VQC Scoring   │ • Claude Sonnet / Opus   │    │  │
│  │  │              │ • VQC Features  │ • Gemini 2.5 Flash       │    │  │
│  │  │              │ • Grover Search │ • GPT-4o                 │    │  │
│  │  │              │ (PennyLane)     │ • Offline Fallback       │    │  │
│  │  └──────────────┴─────────────────┴──────────────────────────┘    │  │
│  └────────────────────────────────────────────────────────────────────┘  │
│                                                                         │
│  ┌────────────────────────────────────────────────────────────────────┐  │
│  │  GenomicSophia/app.py — Co-Pilot Triage Dashboard (port 5001)     │  │
│  │  ┌───────────────────┬───────────────────┬────────────────────┐   │  │
│  │  │  4-Metric Scoring │  VQS API Client   │  LLM Summaries    │   │  │
│  │  │  (ai_engine.py)   │  (vqs_client.py)  │  (ai_engine.py)   │   │  │
│  │  │                   │                   │                    │   │  │
│  │  │  • ABCD Score     │  • IAM Auth       │  • Claude Sonnet   │   │  │
│  │  │  • ClinVar Score  │  • VQS Query      │  • Gemini Flash    │   │  │
│  │  │  • Community Freq │  • Schema Fetch   │  • GPT-4o          │   │  │
│  │  │  • QA Confidence  │  • CSV Parser     │  • Deterministic   │   │  │
│  │  └───────────────────┴───────────────────┴────────────────────┘   │  │
│  └────────────────────────────────────────────────────────────────────┘  │
│                                                                         │
│  ┌────────────────────────────────────────────────────────────────────┐  │
│  │  variant_analysis_ai.py — CLI Batch Analyzer                      │  │
│  │  • Claude API (Opus 4) batch CSV scoring + 2-sentence summaries   │  │
│  └────────────────────────────────────────────────────────────────────┘  │
│                                                                         │
│  Shared: vqs_client.py · ai_engine.py                                   │
└─────────────────────────────────────────────────────────────────────────┘

🚀 Getting Started

Prerequisites

Python 3.10+
(Optional) An API key for one of: Anthropic Claude, Google Gemini, or OpenAI GPT-4o

Installation

# Clone the repository
git clone https://github.com/yacine-baghli/GenomicOracle.git
cd GenomicOracle

# Create a virtual environment
python -m venv .venv

# Activate it
# Windows
.venv\Scripts\activate
# macOS / Linux
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Run — GenomicOracle (Full Analysis)

python main.py

Open http://127.0.0.1:5000 — enter patient data, upload VCF files or manually enter variants, and generate AI-powered clinical reports.

Run — Genomic Co-Pilot (Triage Dashboard)

cd GenomicSophia
pip install -r requirements.txt
python app.py

Open http://127.0.0.1:5001 — the dashboard auto-loads and ranks all CSV files from samples/ on startup.

Run — CLI Batch Analysis

# Set API key
set ANTHROPIC_API_KEY=sk-ant-...

# Analyze one or more CSV files
python variant_analysis_ai.py samples/patient1.csv samples/patient2.csv --output results.csv

⚙️ Configuration

LLM Provider Setup

You can configure your LLM provider in two ways:

Via the UI — Click ⚙️ API Key in the dashboard header, select your provider, paste your key, and save.
Via environment variables:

# Set one of the following before launching
set ANTHROPIC_API_KEY=your-key     # Anthropic Claude (recommended)
set GEMINI_API_KEY=your-key        # Google Gemini
set OPENAI_API_KEY=your-key        # OpenAI GPT-4o

Note: If no API key is provided, both interfaces operate in Offline Deterministic Mode using built-in expert rule-bases that cover 21+ actionable oncological genes (EGFR, KRAS, BRAF, PIK3CA, ALK, BRCA1/2, RET, NTRK, MET, and more).

🧪 Usage — GenomicOracle (Full Analysis)

1. Enter Patient Data

Fill in the patient's name, age, gender, and clinical condition (e.g., Lung Adenocarcinoma).

2. Provide Genetic Variants

Upload a VCF file, or
Manually enter variants (Gene, HGVS notation, Zygosity)
Or click Load Sample Data to try one of 5 pre-built clinical scenarios

3. Run AI Analysis

The platform will:

Annotate variants against ClinVar
Score pathogenicity with the Variational Quantum Circuit
Search for multi-variant interactions via Grover-inspired quantum search
Generate an LLM-powered clinical oncology report (or expert rule-base fallback)

4. Explore Results

Read targeted therapy recommendations aligned with NCCN/ESMO guidelines
Chat with the AI about the patient's genomic profile
Download the report as a PDF

5. Train the Quantum Model (optional)

Navigate to the ⚛ Train Quantum Model page to run the VQC training pipeline with live loss convergence visualization.

🧪 Usage — Genomic Co-Pilot (Triage Dashboard)

Data Input

Tab	How
📄 CSV Data	Drop SOPHiA DDM™ export CSVs into `GenomicSophia/samples/` — they auto-load. Or upload via the dashboard.
🔌 VQS API	Enter SOPHiA DDM™ credentials + dataset keys from browser DevTools.

Ranking Workflow

Load data — CSVs from /samples are auto-loaded on page load
Rank cases — Click 🔬 Rank All Cases (or 🔍 Analyze for VQS API)
Review — Cases are sorted by urgency. Click any case card to expand it.
Explore variants — The variant table shows all scored variants, sorted by composite score
Literature — Click 📚 on any variant row to search PubMed for relevant papers
AI Summaries — Set an LLM API key and select a provider to get AI-generated clinical summaries

What You See Per Case

4-metric radar — ABCD Prediction, ClinVar Evidence, Community Frequency, QA Confidence
🧠 AI Clinical Summary — 2-sentence LLM narrative (when API key is configured)
⚡ Deterministic Analysis — Always-available 2-sentence summary + bullet breakdown + action label
💊 Therapy tags — Matched from the 21-gene actionable database
Variant table — Full priority ranking with gene links to ClinVar, per-variant PubMed search

🔬 Scoring System — 4 Metrics (Co-Pilot)

Variant-Level Scoring

Each variant is scored across 4 metrics (0–100), weighted and combined into a composite score.

1. ABCD Prediction (Weight: 35%)

SOPHiA DDM™'s proprietary ABCD predictor — the primary pathogenicity signal.

ABCD Class	Base Score	+ Consequence Bonus
A — High confidence pathogenic	65	Nonsense/Frameshift/Splice: +30
B — Moderate confidence	45	Missense: +12
C — Low confidence	20	Inframe: +5
D — Benign prediction	5	—

2. ClinVar Evidence (Weight: 25%)

ClinVar Significance	Points	+ Review Status	Points
Pathogenic	+55	Expert panel	+35
Likely Pathogenic	+40	Multiple submitters	+25
VUS	+15	Single submitter	+15
Benign	+5	No assertion	+5

3. Community Frequency (Weight: 20%)

SOPHiA DDM™ lab-wide frequency — rarer variants score higher.

Community Freq.	Score	Interpretation
Absent / 0	95	Never seen across all labs
< 0.1%	90	Ultra-rare
< 0.5%	75	Rare
< 1%	60	Low frequency
≥ 10%	10	Likely artifact

4. QA Confidence (Weight: 20%)

Sequencing quality — starts at 50, adjusted by read depth, VAF, and ABCD.

Factor	Condition	Adjustment
Read Depth	≥ 200×	+25
VAF	≥ 30%	+20
VAF	< 5%	−10
ABCD	A	+10

Case-Level Ranking

Case Score = Top Variant × 0.50 + Average Score × 0.30 + High-ABCD Bonus × 0.20

Case Score	Tier	Action
≥ 70	🔴 CRITICAL	Immediate review required
≥ 50	🟠 HIGH	Priority review
≥ 30	🟡 MODERATE	Standard review
< 30	🟢 LOW	Routine processing

💊 Actionable Gene Database

The platform maintains a curated database of 21 actionable genes with Tier 1/2 classifications and matched FDA-approved therapies:

Gene	Tier	Matched Therapies
EGFR	1	Osimertinib, Erlotinib, Gefitinib
ALK	1	Alectinib, Lorlatinib, Crizotinib
BRAF	1	Dabrafenib+Trametinib, Vemurafenib
BRCA1/2	1	Olaparib, Rucaparib
ROS1	1	Crizotinib, Entrectinib
RET	1	Selpercatinib, Pralsetinib
NTRK1/2/3	1	Larotrectinib, Entrectinib
KRAS	2	Sotorasib (G12C), Adagrasib (G12C)
MET	2	Capmatinib, Tepotinib
HER2/ERBB2	2	Trastuzumab deruxtecan
PIK3CA	2	Alpelisib
ESR1	2	Elacestrant
FGFR2/3	2	Pemigatinib, Erdafitinib
IDH1/2	2	Ivosidenib, Enasidenib

Additionally, 6 prognostic markers are tracked: TP53, PTEN, RB1, CDKN2A, APC, SMAD4.

⚛️ Quantum ML Components

Variational Quantum Circuit (VQC) Scoring

A PennyLane-based quantum circuit (default.qubit simulator) that:

Encodes variant features into quantum states via RX and RY rotations
Outputs a pathogenicity score (0–1) from PauliZ expectation values
Supports post-training VQC feature extraction with interpretable variant topology descriptions

Grover-Inspired Combinatorial Search

Simulates quantum amplitude amplification across multi-variant state spaces to identify clinically significant co-occurring mutations:

Interaction	Genes	Clinical Impact
Synergistic Resistance	KRAS + PIK3CA	Combined MAP/PI3K pathway hyper-activation
Negative Prognostic	EGFR + TP53	Severely limits TKI response duration
Synthetic Lethality Bypass	BRAF + PTEN	PTEN loss bypasses BRAF inhibition
Synthetic Lethality Synergy	PIK3CA + BRCA2	Combined PARP + PI3K inhibition response

📜 Sample Clinical Scenarios

GenomicOracle — 5 Built-in Patients

Patient	Condition	Key Variants
Jane Smith, 62F	Stage IV NSCLC	EGFR L858R, EGFR T790M, TP53 R273H
Michael Johnson, 58M	Metastatic CRC	KRAS G12V, PIK3CA E545K, APC R876*
Sarah Williams, 45F	Stage III Melanoma	BRAF V600E, PTEN R130G, CDKN2A R58*
Emily Davis, 40F	Metastatic Breast (HR+/HER2-)	PIK3CA H1047R, ESR1 D538G, BRCA2 S1982fs
David Brown, 8M	High-Risk Neuroblastoma	MYCN Amplification, ALK F1174L

Genomic Co-Pilot — 8 SOPHiA DDM™ CSV Samples

8 real-format SOPHiA DDM™ export CSVs are included in GenomicSophia/samples/ (patient1–patient8), containing 125–500+ variants each with full ABCD predictions, ClinVar annotations, community frequencies, and read depth data.

📁 Project Structure

GenomicOracle/
├── main.py                    # Flask app — full patient analysis (port 5000)
├── qml_module.py              # Quantum ML: VQC scoring, features, Grover search
├── llm_module.py              # Multi-LLM: treatment reports, chat, offline fallback
├── ai_engine.py               # 4-metric scoring engine + CSV parser + LLM summaries
├── vqs_client.py              # SOPHiA VQS API client (auth + query + schema)
├── variant_analysis_ai.py     # CLI batch analyzer (Claude API)
├── requirements.txt           # Python dependencies
├── templates/
│   ├── index.html             # Landing page — patient & variant input
│   ├── analysis.html          # Variant analysis results
│   ├── results.html           # Full clinical report + AI chat
│   ├── results_pdf.html       # PDF-optimized report template
│   ├── training.html          # VQC training dashboard
│   └── copilot.html           # Co-Pilot dashboard (v1)
├── static/
│   └── style.css              # UI stylesheet (glassmorphism dark theme)
├── uploads/                   # Uploaded VCF files (gitignored)
│
├── GenomicSophia/             # ← Co-Pilot Triage Dashboard (standalone)
│   ├── app.py                 # Flask routes & API endpoints (port 5001)
│   ├── ai_engine.py           # 4-metric scoring + deterministic summaries
│   ├── vqs_client.py          # SOPHiA VQS API client
│   ├── requirements.txt       # Minimal dependencies (flask, requests)
│   ├── samples/               # 8 SOPHiA DDM™ CSV exports (auto-loaded)
│   │   ├── patient1.csv       # ... through patient8.csv
│   │   └── ...
│   ├── templates/
│   │   └── copilot.html       # Dashboard HTML
│   └── static/
│       ├── style.css          # Dark/Light theme CSS
│       └── app.js             # Frontend logic
│
└── GenomicOracle/             # ← Earlier standalone version (separate repo)
    └── ...

🔬 API Endpoints

GenomicOracle (`main.py` — port 5000)

Method	Endpoint	Description
`GET`	`/`	Landing page — patient & variant input
`POST`	`/analysis`	Run variant analysis (VCF or manual)
`GET`	`/results`	View clinical report + AI chat
`POST`	`/api/chat`	Interactive AI chatbot (context-aware)
`GET`	`/training`	VQC training dashboard
`POST`	`/api/train_model`	Trigger quantum model training
`GET`	`/download_pdf`	Download clinical report as PDF
`POST`	`/settings`	Save LLM provider + API key
`GET`	`/copilot`	Co-Pilot dashboard (embedded)
`POST`	`/api/copilot/analyze`	VQS API analysis
`POST`	`/api/copilot/csv`	CSV file analysis

Genomic Co-Pilot (`GenomicSophia/app.py` — port 5001)

Method	Endpoint	Description
`GET`	`/`	Dashboard
`GET`	`/api/samples`	List CSV files in `/samples`
`POST`	`/api/upload`	Upload & persist CSVs to `/samples`
`POST`	`/api/remove_sample`	Remove a CSV from `/samples`
`POST`	`/api/rank`	Score & rank all cases (+ optional LLM)
`POST`	`/api/analyze`	Analyze via VQS API
`POST`	`/api/set_key`	Store LLM API key in session
`POST`	`/api/variant_papers`	PubMed paper search for a gene/variant

🛠️ Tech Stack

Layer	Technology
Backend	Flask (Python 3.10+)
Quantum ML	PennyLane (`default.qubit` simulator)
LLM Integration	Anthropic Claude, Google Gemini, OpenAI GPT-4o
VCF Parsing	vcfpy
PDF Generation	xhtml2pdf
SOPHiA DDM™	VQS API (DuckDB-as-a-service) + CSV exports
Literature	PubMed E-utilities (free NCBI API)
Frontend	HTML5, CSS3 (glassmorphism), vanilla JavaScript, Inter font
Charts	Chart.js (training dashboard)

Dependencies

flask>=3.0
pennylane>=0.39
vcfpy>=0.13
xhtml2pdf>=0.2
google-genai>=1.0
openai>=1.0
anthropic>=0.30
requests>=2.31
pandas>=2.0

🔌 SOPHiA DDM™ VQS API

The platform integrates with SOPHiA Genetics' Variant Query Service (VQS) — a DuckDB-powered API for querying genomic datasets stored on the SOPHiA DDM™ platform.

Authentication

POST https://iam-vandv.sophiagenetics.com/account/token
Body: {"username": "...", "password": "..."}
→ Returns: {"access_token": "Bearer ..."}

Querying Variants

POST https://platform-vandv1.sophiagenetics.com/api/variant/query
Params: key=<dataset_key>&engine.paginate=true
Body: {"columns": ["*"], "pagination": {"offset": 0, "limit": 500}}

How to Get Dataset Keys

Log into SOPHiA DDM™ platform
Open browser DevTools → Network tab
Navigate to a patient's variant list
Copy the key parameter from the /api/variant/query request
Paste into the VQS API tab in the dashboard

⚠️ Disclaimer

This platform is a hackathon prototype built for educational and demonstration purposes. It is not intended for clinical use. All treatment recommendations should be validated by qualified medical professionals. The quantum ML components use simulated quantum circuits on classical hardware.

📄 License

This project is open-source under the MIT License.

👥 Team

Built with ❤️ at ETH Zurich for the SOPHiA Genetics × ETH Zurich Future of Health Hackathon 2026

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
GenomicOracle		GenomicOracle
GenomicSophia		GenomicSophia
samples		samples
static		static
templates		templates
version2		version2
.gitignore		.gitignore
README.md		README.md
Sophia ETH - Hackathon Slides.pdf		Sophia ETH - Hackathon Slides.pdf
ai_analysis.csv		ai_analysis.csv
ai_engine.py		ai_engine.py
llm_module.py		llm_module.py
main.py		main.py
qml_module.py		qml_module.py
requirements.txt		requirements.txt
variant_analysis_ai.py		variant_analysis_ai.py
vqs_client.py		vqs_client.py

Folders and files

Latest commit

History

Repository files navigation

🧬 GenomicOracle — AI × Quantum Genomic Analysis Platform

📋 Overview

✨ Key Features

GenomicOracle — Full Patient Analysis

Genomic Co-Pilot — Clinical Triage Dashboard

CLI — Batch Variant Analysis

🏗️ Architecture

🚀 Getting Started

Prerequisites

Installation

Run — GenomicOracle (Full Analysis)

Run — Genomic Co-Pilot (Triage Dashboard)

Run — CLI Batch Analysis

⚙️ Configuration

LLM Provider Setup

🧪 Usage — GenomicOracle (Full Analysis)

1. Enter Patient Data

2. Provide Genetic Variants

3. Run AI Analysis

4. Explore Results

5. Train the Quantum Model (optional)

🧪 Usage — Genomic Co-Pilot (Triage Dashboard)

Data Input

Ranking Workflow

What You See Per Case

🔬 Scoring System — 4 Metrics (Co-Pilot)

Variant-Level Scoring

1. ABCD Prediction (Weight: 35%)

2. ClinVar Evidence (Weight: 25%)

3. Community Frequency (Weight: 20%)

4. QA Confidence (Weight: 20%)

Case-Level Ranking

💊 Actionable Gene Database

⚛️ Quantum ML Components

Variational Quantum Circuit (VQC) Scoring

Grover-Inspired Combinatorial Search

📜 Sample Clinical Scenarios

GenomicOracle — 5 Built-in Patients

Genomic Co-Pilot — 8 SOPHiA DDM™ CSV Samples

📁 Project Structure

🔬 API Endpoints

GenomicOracle (main.py — port 5000)

Genomic Co-Pilot (GenomicSophia/app.py — port 5001)

🛠️ Tech Stack

Dependencies

🔌 SOPHiA DDM™ VQS API

Authentication

Querying Variants

How to Get Dataset Keys

⚠️ Disclaimer

📄 License

👥 Team

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

GenomicOracle (`main.py` — port 5000)

Genomic Co-Pilot (`GenomicSophia/app.py` — port 5001)

Packages