💊 Drug Safety Intelligence System

AI-Powered Pharmacovigilance & Risk Assessment Platform

🌟 Overview • ✨ Features • 📸 Screenshots • 🛠️ Tech Stack • 📦 Installation • 👥 Team

---

Transform raw clinical data into proactive patient safety insights ✨

---

🌟 Overview

The Drug Safety Intelligence System is an end-to-end, AI-powered pharmacovigilance platform designed to analyze adverse drug events and proactively predict patient harm severity and specific reactions. Utilizing real-world clinical data, the system assists healthcare professionals and clinical researchers in early signal detection and risk assessment.

🏥 FDA FAERS Integration - Built on massive real-world relational databases (Year 2025) including DEMO, DRUG, REAC, THER, and OUTC.

🎯 What Makes This System Special?

🧠 Advanced ML Models Utilizes highly optimized XGBoost models for both Binary Severity Prediction (ROC-AUC: 85.08%) and Multi-class Reaction Forecasting. ⚖️ Class Imbalance Handling Programmatically tackles real-world clinical data imbalance using strategic downsampling and rigorous Target/Mean Encoding strategies. 📊 Global Safety Signals Mines integrated clinical databases to identify and flag statistically significant drug-adverse event pairs automatically.

☁️ Cloud-Optimized Architecture Refactored from a massive 2GB ensemble to a streamlined ~53MB joblib-compressed model, perfect for strict memory environments. 🚀 Dual-Deployment Features both a beautiful, interactive Streamlit Dashboard for end-users and a robust FastAPI Backend for system integrations. 🛠️ Automated SQL Pipeline Constructs complex feature engineering matrices directly using SQL Common Table Expressions (CTEs) without overwhelming RAM.

---

✨ Features

📱 Core Modules

🔬 Patient Risk Analysis Calculates real-time severity probabilities based on patient profiles	🔮 Reaction Forecasting Predicts the top 3 adverse events a patient might experience	🚨 Signal Detection Interactive database exploring historical drug-reaction reports
⚡ Oriva PV API FastAPI endpoints for seamless frontend/external integration	⚙️ Dual Pipeline Distinct, rigorous preprocessing strategies for different models	📈 Deep EDA Reports High-res clinical pattern visualizations and data drift checks

🤖 Machine Learning Capabilities

Task	Algorithm	Optimization	Performance (Test Set)
🔴 Severity Prediction	XGBClassifier (Binary)	scale_pos_weight=1.40, tree_method='hist'	ROC-AUC: 85.08% Recall: 77.00%
💊 Reaction Forecasting	XGBClassifier (Multi-class)	max_depth=10, compress=9 (~53MB footprint)	Accuracy: 72.85% (Across 20 targets)

---

📸 Screenshots

💻 Live System Dashboard

Patient Risk Analysis (Matrix A & B)	Global Safety Signals Database

📊 Deep Exploratory Data Analysis (EDA)

Demographics (Age/Weight)	Target Imbalance	Top 10 Suspect Drugs
Polypharmacy Impact	Age by Gender	Feature Importance
Therapy Duration Analysis
Temporal Stability (Data Drift)

---

🛠️ Tech Stack

Data Processing & Pipeline

Technology	Purpose
	Core Programming Language (3.9+)
	Clinical Data Manipulation
	Mathematical Operations
	Relational Database Engine

Machine Learning

Technology	Purpose
	Preprocessing, Resampling, Metrics
	Tree-based Classification Algorithms
	Hyperparameter Tuning
	Advanced Model Compression

Deployment & Backend

Technology	Purpose
	High-performance RESTful API
	Interactive Web Dashboard
	CI/CD Pipeline (Linting)

---

📦 Installation

📋 Prerequisites

• ✅ Python 3.9 or higher
• ✅ Git installed
• ✅ Raw FDA FAERS Dataset (Not included in repo due to size)

🚀 Quick Start

# 1️⃣ Clone the repository
git clone [https://github.com/GoldenBoy13420/Drug-Safety-Intelligence-System.git](https://github.com/GoldenBoy13420/Drug-Safety-Intelligence-System.git)
cd Drug-Safety-Intelligence-System

# 2️⃣ Install dependencies
pip install -r requirements.txt

# 3️⃣ Run the Streamlit Dashboard
streamlit run app.py

# 4️⃣ Or Run the FastAPI Backend
uvicorn api:app --reload

To rebuild the matrices and retrain the models from scratch:

python main.py

This single entry point will sequentially run:

1. Database Connection Validation
2. Deep EDA (eda.py)
3. Safety Signal Detection (signal_detector.py)
4. Matrix A & B Construction (preprocessing.py)
5. XGBoost Model Training & Evaluation

---

📁 Project Structure

DRUG-SAFETY-INTELLIGENCE-SYSTEM/
│
├── 📂 data/                   # Raw FDA data & SQLite DB (Ignored in Git)
├── 📂 logs/                   # System execution logs (pipeline.log)
├── 📂 models/                 # Compressed XGBoost .pkl models & Encoders
├── 📂 notebooks/              # Data integration & experimental notebooks
├── 📂 reports/
│   ├── 📂 figures/            # High-res EDA plots & Dashboard Screenshots
│   └── 📄 safety_signals_report.csv
│
├── 💻 src/                    # Core Modules
│   ├── config.py              # System paths and constants
│   ├── data_pipeline.py       # Database validation
│   ├── eda.py                 # Visual analysis generator
│   ├── preprocessing.py       # SQL CTEs and Matrix Building
│   ├── model_severity.py      # Binary Classification Pipeline
│   ├── model_adverse_event.py # Multi-class Classification Pipeline
│   └── signal_detector.py     # Pharmacovigilance Mining
│
├── ⚡ api.py                  # FastAPI endpoints
├── 🌐 app.py                  # Streamlit UI dashboard
├── 🚀 main.py                 # Full Pipeline orchestrator
└── 📄 requirements.txt        # Cloud dependencies

---

👥 Team

Mahmoud Abdelrauf 🎯 AI Engineer / Lead

Ahmed Elabd 💻 Data Scientist

Abdelrhman Batran 💻 Data Scientist

📜 License

This project is licensed under the MIT License.

---

⭐ Support

If you find this intelligence system helpful, please give it a star!