cagqsar 1.0.0

A complete CLI QSAR pipeline for drug discovery and predictive toxicology

CAG-QSAR CLI Tool

A complete, production-grade command-line QSAR (Quantitative Structure-Activity Relationship) modeling pipeline. This tool automates the process of data curation, molecular descriptor calculation, feature selection, data splitting, model building, and rigorous validation.

---

Installation

You can install this package in Linux/WSL using three different methods, depending on your environment.

Prerequisites

Make sure you have Python (version >= 3.8) and pip installed:

sudo apt update
sudo apt install python3 python3-pip python3-venv -y

Option A: Install from PyPI (Once Published)

After publishing the package to PyPI, you can create a virtual environment and install the tool globally or locally using pip:

# 1. Create a virtual environment
python3 -m venv qsar_env
source qsar_env/bin/activate

# 2. Install the package from PyPI
pip install cagqsar

# 3. Run the CLI tool
cagqsar --help

Option B: Local Source Install (No root access required)

You can build and install the package locally from the repository folder:

# 1. Navigate to the repository directory
cd git_qsar

# 2. Create and activate a virtual environment
python3 -m venv qsar_env
source qsar_env/bin/activate

# 3. Install the package locally
pip install .

# 4. Run the CLI tool
cagqsar --help

Option C: Editable Development Mode

If you plan to modify the source code of the pipeline and want changes to reflect instantly:

pip install -e .

---

Command Line Usage

After installation, run the application from any directory in your terminal:

cagqsar --data <dataset_csv> --smiles <smiles_column> --activity <activity_column> [options]

Core CLI Arguments:

• --data: Path to the CSV dataset (Required).
• --smiles: Column name containing SMILES strings (Required).
• --activity: Column name containing activities in nM (Required).
• --model: Regression algorithm to train: mlr (MLR), pls (PLS), rf (Random Forest), svr (SVM), xgb (XGBoost), or gnn (Graph Neural Network) (Default: pls).
• --split: Splitting method: random or pca (Kennard-Stone PCA-distance split) (Default: pca).
• --test_size: Fraction of data allocated to the test set (Default: 0.2).
• --var_thresh: Variance filter threshold for dropping constant descriptors (Default: 0.01).
• --corr_thresh: Correlation threshold for collinearity filter (Default: 0.85).
• --y_rand_runs: Number of Y-randomization validation loops (Default: 50).
• --fingerprints: Flag to compute 2D fingerprints (Morgan/ECFP + MACCS keys) in addition to physical descriptors.
• --out_dir: Directory to export curated data, model reports, trained model binaries, and evaluation plots (Default: qsar_output).

---

Programmatic Import in Python

Once the package is installed, you can import and use any of its internal logic (like the structure curator or descriptor calculator) in your own scripts:

from cagqsar import curate_molecule, get_rdkit_descriptors

# 1. Clean a SMILES structure and remove salt fragments
clean_smiles, mol = curate_molecule("CN(C)C(=O)c1ccccc1.Cl", Chem.SaltRemover.SaltRemover())

# 2. Extract standard RDKit descriptors
descriptors = get_rdkit_descriptors(mol)

---

Publishing to GitHub & PyPI

Follow these instructions to publish your code for public access.

1. Publishing to GitHub

Initialize the local git repository, commit the files, and push to GitHub:

# 1. Initialize repository
git init

# 2. Add files (automatically respects .gitignore)
git add .

# 3. Create initial commit
git commit -m "feat: initial release of cagqsar v1.0.0"

# 4. Set main branch name
git branch -M main

# 5. Add remote GitHub link and push
git remote add origin https://github.com/YOUR_USERNAME/cagqsar.git
git push -u origin main

2. Publishing to PyPI

To make the tool installable globally via pip install cagqsar, build and upload the package distributions to the Python Package Index (PyPI):

# 1. Install packaging build tools
pip install --upgrade build twine

# 2. Compile source distribution (sdist) and binary wheel (bdist_wheel)
python3 -m build

# 3. Verify build files
twine check dist/*

# 4. Upload to PyPI (requires PyPI API Token)
python3 -m twine upload dist/*

---

Acknowledgments & Credits

• Concept, Idea & Planning: Sathish Kumar M Ponnaiya (SKM Ponnaiya).
• Infrastructure & Support: Ponnaiya's Code And Genome Pvt Ltd, Madurai (System support, server resources, internet facilities, and infrastructure).
• AI Coding Partner: Pair-programmed and optimized using Antigravity, a Google DeepMind agentic coding system.
• Large Language Model (LLM): Driven by Google's Gemini 3.5 Flash.
• Access Provider: Grateful to Jio for enabling Gemini Premium access.
• Test Dataset: Sourced from the public BindingDB database.

This software is open-access and free for all users under the terms of the MIT License.