🧬 PCV2 Epitope Mapping & Prediction Platform

A structure-aware, machine learning–driven bioinformatics pipeline for predicting antibody-accessible epitopes on the Porcine Circovirus Type 2 (PCV2) capsid protein (ORF2).

---

🌐 Live Application

🔗 https://pcv2.epitope.aiconceptlimited.com.ng/

---

🎯 Overview

This project implements a research-grade computational framework integrating:

• Evolutionary sequence analysis
• Structural biology (PDB-based features)
• Physicochemical characterization
• Machine learning (XGBoost)

to identify potential B-cell epitopes on the PCV2 capsid protein.

💡 Applications

• Epitope discovery
• Vaccine target identification
• Viral antigen characterization
• Immunoinformatics research

---

🧠 Core Concept

Epitope prediction is treated as a multi-modal biological inference problem:

Sequence → Evolution → Structure → Features → ML → Prediction → Validation

---

🔬 Data Sources

Component	Source
Protein sequences	NCBI (Entrez API)
Reference sequence	UniProt
Protein structures	PDB (3R0R, 6EZG)
Epitope validation	IEDB

---

⚙️ Pipeline Architecture

NCBI Retrieval
     ↓
Sequence Cleaning (capsid-only filtering)
     ↓
Multiple Sequence Alignment (MAFFT)
     ↓
Feature Engineering
  - Conservation
  - Entropy
  - SASA
  - Residue Depth
  - Electrostatics
     ↓
Feature Matrix Construction
     ↓
Epitope Labeling (IEDB)
     ↓
Machine Learning (XGBoost)
     ↓
Prediction
     ↓
3D + Sequence Visualization (Streamlit)

---

🧬 Feature Engineering

🧬 Evolutionary Features

• Conservation score (frequency-based)
• Shannon entropy (sequence variability)

🧱 Structural Features

• Solvent Accessible Surface Area (SASA)
• Residue depth
• Secondary structure (loop/helix/sheet)
• Electrostatics

⚗️ Physicochemical Features

• Hydrophobicity
• Charge distribution

🔄 Contextual Features

• Sliding window (±2 residues)
• Spatial neighborhood aggregation

---

🤖 Machine Learning

• Model: XGBoost Classifier
• Input: Residue-level feature matrix
• Output: Probability of epitope per residue

Training Strategy

• Imbalanced dataset handling
• Threshold tuning (default: 0.25)
• Feature importance extraction

---

📊 Results

Metric	Value
Total residues	~162–245
Predicted epitopes	~24
Validated (IEDB overlap)	~4
ROC-AUC	~0.70–0.75

---

🧪 Validation Strategy

• Predictions compared with IEDB experimental epitopes
• Overlap analysis performed at residue level

Interpretation

• ✅ Overlapping residues → validated epitopes
• 🔬 Non-overlapping → novel candidate epitopes

---

🧬 Biological Insights

Predicted epitopes are enriched in:

• Surface-exposed regions (high SASA)
• Loop/coil structures
• High-entropy (variable) regions

👉 This aligns with known principles of antibody binding.

---

📁 Project Structure

pcv2_epitope_project/
│
├── data/                  # Metadata, mappings, IEDB data
├── sequences/             # FASTA + alignments
├── structures/            # PDB files (3R0R, 6EZG)
├── features/              # Engineered features
├── results/               # Predictions + evaluation
├── models/                # Trained ML model
├── scripts/               # Feature + analysis scripts
├── pipeline/              # Automation scripts
│
├── dashboard.py           # Streamlit interface
└── run_smart_pipeline.py  # Full pipeline runner

---

⚙️ Installation

git clone https://github.com/YOUR_USERNAME/pcv2-epitope-platform.git
cd pcv2-epitope-platform

python -m venv pcv2_env
source pcv2_env/bin/activate

pip install -r requirements.txt

---

▶️ Usage

Run Full Pipeline

python run_smart_pipeline.py

Launch Dashboard

streamlit run dashboard.py

---

📊 Dashboard Features

• 📈 Epitope probability plots
• 🧬 Sequence visualization (UniProt-aligned)
• 🧊 3D structure mapping (Py3Dmol)
• 🧪 IEDB validation overlay
• 📦 Epitope clustering

---

⚠️ Limitations

• Limited experimentally validated epitopes (class imbalance)
• Predictions are computational (require lab validation)
• Sequence–structure mapping introduces approximation

---

🚀 Future Work

• Graph Neural Networks (GNN)
• Transformer-based protein models
• Improved structural alignment
• REST API deployment
• Continuous data updates (automated pipeline)

---

🤝 Collaboration

Open to collaborations in:

• Bioinformatics
• Immunoinformatics
• Vaccine design
• Structural biology

---

📜 Disclaimer

This system provides computational predictions and should not replace experimental validation.

---

👤 Author

Abubakar Bioinformatics & Computational Biology

---

⭐ Acknowledgements

• NCBI (sequence data)
• RCSB PDB (structural data)
• IEDB (epitope data)
• Biopython, XGBoost, Streamlit communities