TCGPR

A Python library for divide-and-conquer (TCGPR) - an efficient strategy tailored for small datasets in materials science and beyond.

---

📖 Citation

If you use this code in your research, please cite the following papers:

• Li T., Cao B., Su T., ... Feng L., Zhang T. Machine Learning-Engineered Nanozyme System for Synergistic Anti-Tumor Ferroptosis/Apoptosis Therapy, SMALL Link to paper

• Wei Q., Cao B., Yuan H., ... Dong Z., Zhang T. Divide and conquer: Machine learning accelerated design of lead-free solder alloys with high strength and high ductility, npj Computational Materials Link to paper

---

📜 Project History

• 2022: TCGPR was first proposed and implemented, in collaboration with Mr. Hao Yuan (experiments) and Mr. Qinghua Wei (experiments). It was successfully applied to the optimization of lead-free solder alloys. → Published in npj Computational Materials News link

• 2024: After two years of development, TCGPR was enhanced with sequential feature selection and outlier detection. In collaboration with Mr. Tianliang Li (experiments) and Mr. Tianhao Su (computations), it was applied to anti-tumor ferroptosis studies. → Published in SMALL News link

---

🧠 Algorithm Overview

For an in-depth explanation of the algorithm, see the TCGPR Introduction PDF.

---

🔧 Installation

Install TCGPR via PyPI:

bash pip install PyTcgpr

To verify the installation:

bash pip show PyTcgpr

To upgrade to the latest version:

bash pip install --upgrade PyTcgpr

---

🚀 Getting Started

1. Data Screening | Partition Mode

```python from PyTcgpr import TCGPR

TCGPR.fit( filePath = "data.csv", initialsetcap = 3, samplingcap = 2, upsearch = 500, CV = 'LOOCV', Task = 'Partition' ) ```

2. Data Screening | Identification Mode

```python from PyTcgpr import TCGPR

TCGPR.fit( filePath = "data.csv", samplingcap = 2, upsearch = 500, CV = 'LOOCV', Task = 'Identification' ) ```

3. Feature Selection Mode

```python from PyTcgpr import TCGPR

TCGPR.fit( filePath = "data.csv", Mission = 'FEATURE', samplingcap = 2, upsearch = 500, CV = 'LOOCV' ) ```

---

⚙️ Parameters

```python :param Mission: str, default='DATA' - 'DATA': Perform data screening - 'FEATURE': Perform feature selection

:param filePath: str Path to input dataset in CSV format

:param initialsetcap: int or list Initial subset size or index list for Partition mode

:param sampling_cap: int, default=1 Number of items selected per iteration

:param measure: str, default='Pearson' Correlation type: 'Pearson' or 'Determination'

:param ratio: float Tolerance threshold for correlation-based filtering

:param target: int, default=1 Number of targets in regression (for feature selection)

:param weight: float, default=0.2 Weight coefficient in GGMF score calculation

:param up_search: int, default=500 Upper limit for search iterations

:param exploit_coef: float, default=2 Variance constraint for EI acquisition function

:param exploit_model: bool, default=False If True, disables GGMF and uses only R values

:param CV: int or str, default=10 Cross-validation: integer (e.g., 5, 10) or 'LOOCV' ```

---

📤 Output

After running, TCGPR outputs a CSV file with the remaining samples:

bash Dataset_remained_by_TCGPR.csv

---

📦 Source Code

Compatible with Windows, Linux, and macOS.

---

🧾 Patent

---

👨‍🔧 Maintainer

Maintained by Bin Cao 📧 Email: bcao686@connect.hkust-gz.edu.cn Feel free to open an issue or contact me for any questions, bugs, or collaboration opportunities.

---

🤝 Contributing

Contributions and suggestions are welcome!

• Report bugs or request features via GitHub Issues
• Submit a pull request with improvements or fixes
• Interested in research collaboration? Please get in touch!