🧠 trbxai — Trustworthy & Explainable AI for Clinical and Synthetic Health Data (coming soon-Nov2025)

trbxai is a Python package for interpretable analysis of clinical and synthetically generated health data,
combining SHAP-based explainability and GAN-powered synthetic generation.
Designed for clinical researchers, AI developers, and scientists working with imbalanced, rare, or sensitive datasets.

---

✅ Key Features

• 🧬 Synthetic Data Generation (ctGAN)
  Generate new, realistic samples based on real clinical datasets.

• 🧠 Explainability with SHAP
  Visualize the most influential features in your model, per subject or globally.

• 📊 SHAPSet Plot (coming soon)
  Visualize interactions between feature sets using shapiq.

• 💬 Natural Language Output
  Narrative summaries explaining key predictors in plain language.

---

🔧 Installation

Development version:

git clone https://github.com/DanteTrb/trbxai.git

pip install -e .

🚀 Quickstart

1. Generate synthetic patients

from trbxai.generate.ctgan_generator import generate_synthetic_patients synthetic_df = generate_synthetic_patients(real_df, num_samples=100) 2. Explain model predictions

from trbxai.explain.shapset_explainer import get_shap_values shap_values, explainer = get_shap_values(model, X, feature_names=X.columns.tolist())

🧪 Clinical Use Case

You have only 46 patients with hereditary cerebellar ataxia. You want to: Augment the dataset with synthetic patients Train a robust classifier Understand which gait variables predict disease With trbxai, you can do it — in just a few steps.

📁 Project Structure

trbxai/ ├── generate/ # ctGAN module │ └── ctgan_generator.py ├── explain/ # SHAP and explainability module │ └── shapset_explainer.py ├── tests/ # Unit tests with pytest ├── pyproject.toml ├── README.md └── setup.cfg

🧪 Testing

To run all tests: pytest

📘 Docs and Examples

🧪 Example notebooks (coming soon, stay tuned)

🧠 Streamlit app integration (in development)

📊 /examples folder will be available soon

❤️ Contributing

Open an issue, submit a pull request, or suggest new features! If you use trbxai in your research, consider citing the repo (BibTeX coming soon).

📜 License

Distributed under the MIT License.

🔮 Vision

We aim to make trbxai the go-to library for Explainable AI in clinical research, bridging data science and clinical decision-making with tools that are trustworthy, interpretable and deployable.