---
license: apache-2.0
pretty_name: Hippolv ADMET & Solubility Dataset
size_categories:
- 1K<n<10K
task_categories:
- text-generation
- question-answering
language:
- en
tags:
- chemistry
- admet
- pharmacology
- smiles
- openadmet
- medical
- drug-discovery
dataset_info:
  features:
  - name: instruction
    dtype: string
  - name: input
    dtype: string
  - name: output
    dtype: string
---

<div align="center">
  <h1>💊 Hippolv</h1>
  <p><i>A highly curated, instruction-tuning dataset for teaching AI how drugs behave in the human body (ADMET & Solubility).</i></p>
</div>

<p align="center">
  <img src="https://img.shields.io/badge/Dataset_Size-~9.47k_Rows-blue?style=for-the-badge" alt="Size">
  <img src="https://img.shields.io/badge/Format-Apache_Parquet-green?style=for-the-badge" alt="Format">
  <img src="https://img.shields.io/badge/Language-Global_English-purple?style=for-the-badge" alt="Language">
  <img src="https://img.shields.io/badge/License-Apache 2.0-orange?style=for-the-badge" alt="License">
</p>

<hr>

## 💡 Overview

Welcome to **Hippolv**, the third specialized dataset curated by [ZemResearch](https://huggingface.co/ZemResearch). While our previous datasets focused on general molecular structures and toxicology, Hippolv is engineered to bridge the gap between pure chemistry and clinical pharmacology.

This dataset focuses on **ADMET** (Absorption, Distribution, Metabolism, Excretion, and Toxicity) and aqueous solubility. We sourced raw, high-quality clinical experimental data from the prestigious **OpenADMET Challenges** (`expansionrx` and `pxr-challenge`), and transformed them into a clean, instruction-following format. It's the perfect fuel to fine-tune your Chemical LLMs into expert pharmaceutical assistants.

## 🧼 The "Deep Clean" Sterilization Process

High-quality models require high-quality data. We didn't just scrape this data; we rigorously sterilized it using a strict RDKit pipeline:
1. **Chemical Validation:** Every molecule was passed through RDKit. If the SMILES string was broken or chemically invalid, it was instantly dropped.
2. **Canonicalization:** All valid molecules were converted into their standard **Canonical SMILES** (with isomeric properties preserved) to ensure structural consistency.
3. **Deduplication:** We aggressively dropped duplicates based on their true chemical structure, not just string similarity. We also removed any rows with missing ADMET values.
4. **Result:** A hyper-sterilized dataset of pure, instruction-ready ADMET data, free from data leakage and benchmark conflicts.

## 📦 Why is the file so tiny?

If you check the files, you might be surprised to see that thousands of rows of pharmaceutical data fit into a file of around **1 MB**. 

This is the magic of the **Apache Parquet** format paired with Google's **Snappy compression**. Since SMILES strings and repetitive text formats compress incredibly well, you get a feather-light file to download. Once loaded into your Python environment, it instantly decompresses to its full size, saving your bandwidth and keeping your GPU data-loading ultra-fast!

## 💻 How to Use (Quick Start)

You don't need to manually download anything. Just use the `datasets` library by Hugging Face to load it directly into your workflow:

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("ZemResearch/Hippolv")

# Check the total rows
print(f"Total verified ADMET data: {len(dataset['train'])}")

# Look at the first data point
print(dataset['train'][0])
```

## 📝 Data Structure

Hippolv is formatted for standard Instruction-Tuning (Alpaca/ChatML style). Each row contains three columns:

* `instruction`: The prompt asking the AI to analyze the molecule's properties.
* `input`: The verified Canonical SMILES string of the molecule.
* `output`: A structured narrative detailing the compound's ADMET and solubility metrics.

### Example Row:

```json
{
  "instruction": "Analyze the aqueous solubility and ADMET profile for the following chemical compound.",
  "input": "CC(=O)Nc1ccc(O)cc1",
  "output": "ADMET Metrics:\n- LogD: 0.46\n- Kinetic Solubility (KSOL): High\n- HLM CLint: 12.5\n- Caco-2 Permeability: Good"
}
```

*(Note: The example output reflects the combined narrative format generated during our preprocessing pipeline).*

## 🤝 Citation & Collaboration

Created with ❤️ by [ZemResearch](https://huggingface.co/ZemResearch). If you use Hippolv for your academic research, drug discovery pipelines, or to train your own AI models, we'd love to hear about it! Feel free to open a discussion in the community tab.