klimczakjakubdev/pharmaco-explainer

---
license: cc-by-4.0
language:
- en
tags:
- chemistry
configs:
- config_name: k3
default: true
data_files:
- "k3/k3_split.parquet"

- config_name: k4
data_files:
- "k4/k4_split.parquet"

- config_name: k4_2ar
data_files:
- "k4_2ar/k4_2ar_split.parquet"

- config_name: k5
data_files:
- "k5/k5_split.parquet"

- config_name: k3_labels
data_files:
- "k3/k3_labels.parquet"

- config_name: k4_labels
data_files:
- "k4/k4_labels.parquet"

- config_name: k4_2ar_labels
data_files:
- "k4_2ar/k4_2ar_labels.parquet"

- config_name: k5_labels
data_files:
- "k5/k5_labels.parquet"
---
# Pharmaco-Explainer Datasets

This repository contains **datasets used in the Pharmaco-Explainer project**.
They are shared separately on Hugging Face and are used by the training and
experimentation code hosted on GitHub:

👉 **Training code and scripts:**
https://github.com/AdamSulek/pharmaco-explainer/

---

## Available Datasets

The following datasets are available:
- **k3:** 3-element pharmacophore
- **k4_2ar:** 4-element pharmacophore with two aromatic features
- **k4:** 4-element pharmacophore
- **k5** 5-element pharmacophore

Each dataset consists of three files:

- `<dataset>.parquet` – main feature data
- `<dataset>_split.parquet` – predefined train/val/test split
- `<dataset>_labels.parquet` – labels

Repository structure:
```
k3/
├── k3.parquet
├── k3_split.parquet
└── k3_labels.parquet
k4_2ar/
├── k4_2ar.parquet
├── k4_2ar_split.parquet
└── k4_2ar_labels.parquet
k4/
├── k4.parquet
├── k4_split.parquet
└── k4_labels.parquet
k5/
├── k5.parquet
├── k5_split.parquet
└── k5_labels.parquet
```

## Downloading the Data

Datasets can be downloaded **directly from Hugging Face** using the Python
script below.

### Requirements

- Python ≥ 3.8
- `requests`
- Environment variable `PHARM_PROJECT_ROOT` pointing to the project root

```bash
import argparse
from pathlib import Path
import requests
import os

def project_path(*parts):
root = os.environ.get("PHARM_PROJECT_ROOT")
if root is None:
raise RuntimeError(
"Environment variable PHARM_PROJECT_ROOT is not set.\n"
"Run:\n"
" export PHARM_PROJECT_ROOT=/path/to/project"
)
return os.path.join(root, *parts)

def download_file(url, dest_path: Path):
dest_path.parent.mkdir(parents=True, exist_ok=True)
if dest_path.exists():
print(f"File already exists, skipping: {dest_path}")
return
print(f"Downloading {url} -> {dest_path}")
r = requests.get(url, stream=True)
r.raise_for_status()
with open(dest_path, "wb") as f:
for chunk in r.iter_content(chunk_size=1024 * 1024):
if chunk:
f.write(chunk)

def download_dataset(dataset_name):
base_url = (
f"https://huggingface.co/datasets/"
f"klimczakjakubdev/pharmaco-explainer/resolve/main/{dataset_name}/"
)
files = [
f"{dataset_name}.parquet",
f"{dataset_name}_split.parquet",
f"{dataset_name}_labels.parquet",
]

dest_dir = Path(project_path("data", dataset_name))
for file_name in files:
url = base_url + file_name
dest_path = dest_dir / file_name
download_file(url, dest_path)

if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"--dataset",
type=str,
default="k3",
choices=["k3", "k4", "k4_2ar", "k5"],
help="Which dataset to download"
)
args = parser.parse_args()

download_dataset(args.dataset)
```

Example Usage
```
python download_dataset.py --dataset k4
```

Files will be downloaded to:
```
$PHARM_PROJECT_ROOT/data/k4/
```

## Related Repository
These datasets are used by the main Pharmaco-Explainer codebase:

🔗 https://github.com/AdamSulek/pharmaco-explainer/

That repository contains:

* model training code
* experiment configurations
* preprocessing scripts
* evaluation and explainability pipelines