QSAR-DENV: Anti-Dengue Drug Discovery from Colombian Medicinal Flora

Overview

This repository contains the complete computational pipeline for QSAR (Quantitative Structure-Activity Relationship) modeling and virtual screening of anti-dengue compounds derived from Colombian medicinal flora. The study integrates ethnobotanical knowledge with machine learning to identify promising drug candidates.

Key Features

• Curated Dataset: 358 Colombian medicinal plants with antiviral activity
• Machine Learning: XGBoost-based QSAR model (MCC = 0.583)
• Virtual Screening: 3,267 phytochemicals screened
• Comprehensive Analysis: Chemical diversity, applicability domain, and SAR analysis
• Bayesian Optimization: Automated hyperparameter tuning using Optuna

Citation

If you use this code or data, please cite:

[citation will be added here once published]

Repository Structure

QSAR-DENV/
├── README.md                          
├── requirements.txt                   
├── environment.yml                   
├── LICENSE                           
├── scripts/                          
│   ├── 01_exploratory_data_analysis.py
│   ├── 02_chemical_diversity_analysis.py
│   ├── 03_data_preparation_baseline_models.py
│   ├── 04_advanced_model_optimization.py
│   └── 05_qsar_prediction_tool.py
├── data/                            
│   └── example_compounds.csv
├── models/                          
│   └── README.md
├── notebooks/                       
│   └── example_usage.ipynb
└── docs/                           
    └── user_guide.md

Installation

Prerequisites

• Python 3.8 or higher
• pip or conda package manager

Option 1: Using pip

# Clone the repository
git clone https://github.com/Sergio111999/QSAR-DENV.git
cd QSAR-DENV

# Install dependencies
pip install -r requirements.txt

Option 2: Using conda

# Clone the repository
git clone https://github.com/Sergio111999/QSAR-DENV.git
cd QSAR-DENV

# Create conda environment
conda env create -f environment.yml
conda activate qsar-denv

Quick Start

1. Exploratory Data Analysis

python scripts/01_exploratory_data_analysis.py

Performs comprehensive EDA on the anti-dengue bioactivity dataset including:

• Activity distribution analysis
• Physicochemical property correlations
• Molecular weight and drug-likeness statistics

2. Chemical Diversity Analysis

python scripts/02_chemical_diversity_analysis.py

Analyzes chemical diversity through:

• Molecular scaffolds identification
• Chemical clustering
• Fingerprint similarity analysis
• Pharmacophore feature extraction

3. Model Training and Optimization

python scripts/04_advanced_model_optimization.py

Trains and optimizes QSAR models using:

• Multiple machine learning algorithms (XGBoost, LightGBM, ExtraTrees)
• Bayesian hyperparameter optimization (Optuna)
• 5-fold cross-validation
• Comprehensive performance metrics

4. Virtual Screening

python scripts/05_qsar_prediction_tool.py --input your_compounds.csv

Screens new compounds with:

• Activity prediction
• Drug-likeness assessment (QED)
• Applicability domain evaluation

Dependencies

Main dependencies (see requirements.txt for complete list):

• rdkit >= 2022.09.1
• pandas >= 1.5.0
• numpy >= 1.23.0
• scikit-learn >= 1.1.0
• xgboost >= 1.7.0
• lightgbm >= 3.3.0
• optuna >= 3.0.0
• matplotlib >= 3.6.0
• seaborn >= 0.12.0

Data Availability

Training Dataset

The curated anti-dengue bioactivity dataset from ChEMBL is available in the supplementary materials of the associated publication.

Colombian Phytochemical Library

The complete library of 3,267 compounds from 358 Colombian medicinal plants is available in the supplementary materials.

Trained Models

Pre-trained models are available for download:

• Download from Zenodo (will be added)

Usage Example

from rdkit import Chem
import pandas as pd
import pickle

# Load trained model
with open('models/xgboost_final_model.pkl', 'rb') as f:
    model = pickle.load(f)

# Load your compounds
df = pd.read_csv('your_compounds.csv')  # Must have 'SMILES' column

# Make predictions
predictions = model.predict_proba(X)[:, 1]

print(f"Predicted anti-dengue activity: {predictions[0]:.3f}")

Model Performance

The final optimized XGBoost model achieved:

• MCC: 0.583

Performance was evaluated using 5-fold cross-validation on a balanced dataset.

Results

Virtual Screening Results

• Total compounds screened: 3,267
• High-activity predictions (>0.7): 276
• High drug-likeness (QED >0.5): 20
• Top priority candidates: 14 novel compounds

Top Identified Compounds

The top priority compounds identified combine:

• High predicted anti-dengue activity
• Favorable drug-likeness properties (QED)
• Location within model's applicability domain
• Novel chemical scaffolds not present in training data

See supplementary materials for complete results.

Documentation

• User Guide - Detailed usage instructions
• Supplementary Materials - Complete data and results

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.