PAD ML Workflow

A Python package for researchers to explore and analyze Paper Analytical Device (PAD) data, build machine learning models, and develop new analytical methods for pharmaceutical quality testing.

About PADs

Paper Analytical Devices (PADs) are low-cost diagnostic tools designed to verify pharmaceutical authenticity in low-resource settings. When a dissolved drug sample is applied to a PAD card, it produces colorimetric patterns that can be analyzed to determine drug quality and composition.

This package provides programmatic access to PAD image data collected through the PADReader mobile app and enables researchers to:

• Explore historical PAD test data
• Apply and evaluate machine learning models
• Develop new analytical methods
• Build custom ML pipelines for PAD analysis

Installation

pip install pad-analytics

Or install directly from GitHub:

pip install git+https://github.com/PaperAnalyticalDeviceND/pad-analytics.git

For development:

git clone https://github.com/PaperAnalyticalDeviceND/pad-analytics.git
cd pad-analytics
pip install -e .

Debug Mode

By default, the package suppresses technical warnings for a cleaner user experience. To enable debug output:

PAD_DEBUG=1 python your_script.py
# or
export PAD_DEBUG=1
python -c "import pad_analytics as pad; pad.predict(19208, 18)"

Note: You may see libpng error: Read Error messages during prediction - these are harmless warnings from corrupted image data on the server side and do not affect the prediction results.

Quick Start

import pad_analytics as pad

# Explore available projects
projects = pad.get_projects()
print(f"Found {len(projects)} projects")

# Get PAD test cards from a specific project
cards = pad.get_project_cards("ChemoPADNNtraining2024")

# Analyze a specific PAD card
card_data = pad.get_card(card_id=19208)
print(f"Drug tested: {card_data['sample_name'].values[0]}")
print(f"Concentration: {card_data['quantity'].values[0]} %")

# Apply a pre-trained model
actual, prediction = pad.predict(card_id=19208, model_id=18)

Key Features

1. Data Exploration

Access the complete PAD database through the OAS-compliant API:

# List all projects
projects = pad.get_projects()

# Get cards by various criteria
cards = pad.get_project_cards(12) # project id
card = pad.get_card(sample_id=65490) # sample id

# View available ML models
models = pad.get_models()

2. Model Application

Apply pre-trained models to PAD images:

# Neural Network models (for classification)
actual, (drug_name, confidence, energy) = pad.predict(card_id=19208, model_id=16)

# PLS models (for concentration quantification)
actual_conc, predicted_conc = pad.predict(card_id=19208, model_id=18)

3. Visualization

Interactive widgets for Jupyter notebooks:

# Display PAD card with metadata
pad.show_card(card_id=19208)

# Show prediction results
pad.show_prediction(card_id=19208, model_id=18)

# Display multiple cards grouped by drug type
cards_df = pad.get_project_cards("ChemoPADNNtraining2024")
pad.show_grouped_cards(cards_df, group_column='sample_name')

Example Notebook

See notebooks/using_padml_package.ipynb for a comprehensive example of:

• Exploring PAD projects and data
• Applying different model types
• Visualizing results
• Evaluating model performance
• Building custom analysis pipelines

Research Applications

This package supports various research activities:

For Chemistry Researchers

• Analyze PAD performance across different drug formulations
• Evaluate colorimetric response patterns
• Optimize PAD card designs
• Validate new analytical methods

For Computer Science Researchers

• Develop new ML models for PAD analysis
• Compare algorithm performance (NN vs PLS vs custom)
• Implement novel image processing techniques
• Create ensemble methods for improved accuracy

The PAD Workflow

1. Sample Preparation: Dissolve pharmaceutical sample
2. Application: Apply sample to PAD card
3. Reaction: Chemical indicators produce color patterns
4. Imaging: Capture with PADReader mobile app
5. Analysis: ML algorithms interpret patterns
6. Results: Determine drug identity and quality

This package focuses on steps 5-6, providing tools to analyze the collected images and develop better analytical methods.

API Documentation

Core Functions

Function	Description	Returns
get_projects()	List all PAD projects	DataFrame of projects
get_card(card_id)	Get specific card data	Card metadata + image URL
predict(card_id, model_id)	Apply model to card	(actual, prediction)
get_models()	List available models	DataFrame of models
show_card(card_id)	Display card in notebook	Interactive widget

Model Types

Neural Networks (TensorFlow Lite)

• Purpose: Drug identification and multi-class classification
• Output: (predicted_class, probability, energy_score)

PLS (Partial Least Squares)

• Purpose: Concentration quantification
• Output: predicted_concentration (float)

Requirements

• Python >= 3.8
• TensorFlow >= 2.13.0
• OpenCV-Python >= 4.5.0
• NumPy, Pandas, scikit-learn
• ipywidgets (for notebook visualizations)

Contributing

We welcome contributions from both chemistry and computer science researchers! Please see our Contributing Guide.

Citation

If you use this package in your research, please cite:

@software{pad_analytics,
  title = {PAD Analytics: Python Tools for Paper Analytical Device Research},
  author = {Paper Analytical Device Project Team},
  institution = {University of Notre Dame},
  year = {2024},
  url = {https://github.com/PaperAnalyticalDeviceND/pad-analytics}
}

License

MIT License - see LICENSE

Links

• PAD Project Homepage
• PADReader Mobile App
• API Documentation
• OpenAPI Specification
• GitHub Repository

Support

For questions about:

• PAD technology and chemistry: Visit padproject.nd.edu
• Package usage and ML models: Open an issue on GitHub
• API access: Check the API documentation

Security Notice

Note about Keras dependency: This package uses Keras 2.14.0 which has a known security vulnerability (CVE-2024-55459) related to the keras.utils.get_file() function. However, pad-analytics is not affected because we do not use this function in our codebase.

If you are extending this package and need to download files programmatically:

• DO NOT use keras.utils.get_file() with untrusted URLs
• If you must use it, add security measures to validate any downloaded files
• Only download files from trusted sources

We plan to upgrade to Keras 3.8.0+ in a future release to eliminate this dependency concern. See Issue #2 for more details.