# XRepDDA

**Repository Path**: Tomhappy/XRepDDA

## Basic Information

- **Project Name**: XRepDDA
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-02-21
- **Last Updated**: 2026-02-21

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# XRepDDA: An Interpretable Drug-Disease Association Prediction Framework Leveraging Pretrained Chemical Language Models

## File Descriptions

  - **`feature/drug.py`**
    This script handles the extraction of drug features. It utilizes the `smi_ted_light` chemical language model to generate high-dimensional molecular representations from SMILES strings. All necessary code for the model is included in the repository.

  - **`undersample.py`**
    This script is for data preprocessing. It integrates the extracted drug features with disease features and then applies the ALLKNN algorithm to perform undersampling on the negative samples, creating a balanced dataset for training.

  - **`train.py`**
    This script trains the ModernNCA model using a 10-fold cross-validation strategy. It automatically evaluates performance, saves the model from each fold, and stores the best-performing model based on the highest AUC score.

  - **`smi_ted_light/`**
    This directory contains the code for the pre-trained **SMI-TED** model, which is used for encoding drug molecules in `feature/drug.py`.

  - **`data/`**
    This directory contains the three datasets (B-dataset, C-dataset, F-dataset) used in this study. Each sub-directory includes drug information, disease features, and known drug-disease associations.

  - **`results/`**
    This directory stores all outputs from the training process. It includes the trained models for each fold, the best overall model, evaluation metrics (`eval_results.csv`), and a summary log (`summary.out`).

## Usage Workflow

The complete workflow can be executed via the command line. The following example uses the B-dataset:

**1. Extract Drug Features**

```bash
python feature/drug.py
```

**2. Perform Data Undersampling**

```bash
python undersample.py
```

**3. Prepare Training Data**
Copy the resampled data to the path expected by the training script.

```bash
# For Linux / macOS
cp ./sample/B/undersample/resampled_samples.csv ./sample/B/resampled_samples.csv

# For Windows PowerShell
copy .\sample\B\undersample\resampled_samples.csv .\sample\B\resampled_samples.csv
```

**4. Train and Evaluate the Model**

```bash
python train.py
```

The script will run the full 10-fold cross-validation and save all results to the `results/B/` directory.

**5. web server**

You can use XRepDDA directly by accessing the web server: http://www.xrepdda.com