🦠 Protein Structure-Based Virulence Prediction

A working Repo for CSE7850 Project, Predicting bacterial virulence and pathogenicity from protein folding patterns using machine learning algorithms

📁 Repository Structure

GeneratingVirulentData.ipynb: Downloads and processes virulent protein sequences from VFDB, converts RefSeq IDs to UniProt IDs, and retrieves AlphaFold structures.

GeneratingNonVirulentData.ipynb: Collects non-virulent protein sequences from non-pathogenic organisms (e.g., Bacillus subtilis), processes them similarly, and prepares them for feature extraction.

MLModelTraining.ipynb: Loads extracted features, performs preprocessing, and trains machine learning models to classify proteins as virulent or non-virulent.

🧬 Workflow Overview

Data Collection:

• Virulent sequences sourced from VFDB

• Non-virulent sequences from UniProt (e.g., B. subtilis)

Structure Retrieval:

• UniProt IDs mapped to AlphaFoldDB entries

• 3D structure files (.pdb) downloaded automatically

Feature Extraction:

• Secondary structure content (α-helix, β-sheet, coil)

• Hydrogen bond density via DSSP

• Confidence metrics (mean pLDDT)

Model Training:

• Feature matrices combined and labeled (1 = virulent, 0 = non-virulent)

• Classifier trained with stratified sampling and evaluated on a held-out test set

⚙️ Dependencies

Install via conda:

conda create -n virulence-prediction python=3.10
conda activate virulence-prediction
conda install -c conda-forge biopython pandas jupyter scikit-learn
pip install tqdm torch esm biopandas

Also required:

AlphaFoldDB access (public download)

DSSP executable (mkdssp, via conda install -c salilab dssp)

📊 Output

virulent_features.csv / nonvirulent_features.csv: Structural features extracted from PDBs

Trained model and performance metrics (accuracy, F1-score, etc.)

Ready-to-use classifier for predicting new protein sequences

🚀 How to Run

Clone this repo:

git clone https://github.com/BowerH/virulence-structure-prediction.git
cd virulence-structure-prediction

Run the notebooks in order:

• Scripts/GeneratingVirulentData.ipynb
• Scripts/GeneratingNonVirulentData.ipynb
• Scripts/MLModelTraining.ipynb

Github Layout

Bacterial_Virulence_Prediction/
├── Data/
│   ├── structure_features_extracted_clean.csv                     # Non-virulent protein features
│   ├── structure_features_extracted_clean_final.csv               # Virulent protein features
├── Scripts/
│   ├── GeneratingVirulentData.ipynb           # Downloads and processes virulent protein data
│   ├── GeneratingNonVirulentData.ipynb        # Downloads and processes non-virulent protein data
│   ├── Model_Training.ipynb                   # Trains classification models
├── Model/
│   ├── ML_final_ensemble_model.pkl            # Trained ensemble model (e.g. RandomForest + XGBoost)
├── Attachments/
│   └── ML_final_model_comparison.csv          # Model benchmarking results
├── README.md                                  # Project overview and usage instructions
└── .gitignore                                 # Ignore logs, checkpoints, cache files, etc.

👥 Contributors

Name	GitHub
	@sanyabadole
	@BowerH
	@binfwizard
	@rfatma3

📌 Citation

If you use this pipeline in your work, please cite the repository and relevant data sources (VFDB, AlphaFoldDB, UniProt).