---
dataset_info:
  features:
  - name: smiles
    dtype: string
  - name: logP
    dtype: float64
  - name: qed
    dtype: float64
  - name: SAS
    dtype: float64
  - name: canonical_smiles
    dtype: string
  - name: single_bond
    dtype: int64
  - name: double_bond
    dtype: int64
  - name: triple_bond
    dtype: int64
  - name: aromatic_bond
    dtype: int64
  - name: ring_count
    dtype: int64
  - name: R3
    dtype: int64
  - name: R4
    dtype: int64
  - name: R5
    dtype: int64
  - name: R6
    dtype: int64
  - name: R7
    dtype: int64
  - name: R8
    dtype: int64
  - name: R9
    dtype: int64
  - name: R10
    dtype: int64
  - name: R12
    dtype: int64
  - name: R13
    dtype: int64
  - name: R14
    dtype: int64
  - name: R15
    dtype: int64
  - name: R18
    dtype: int64
  - name: R24
    dtype: int64
  splits:
  - name: train
    num_bytes: 61223067
    num_examples: 224568
  - name: validation
    num_bytes: 6784626
    num_examples: 24887
  download_size: 22056296
  dataset_size: 68007693
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
---


# Dataset Card for "ZINC250k"

ZINC250k from [Irwin et al., 2005](https://pubmed.ncbi.nlm.nih.gov/15667143/); taken from [Jo et al., 2022](https://arxiv.org/abs/2202.02514).

Data downloaded from: https://github.com/harryjo97/GDSS.
Additional annotations (bond and ring counts) added using [`rdkit`](https://www.rdkit.org/docs/index.html) library.

## Quick start usage:

```python
from datasets import load_dataset

ds = load_dataset("yairschiff/zinc250k")

# Use `ds['train']['canonical_smiles']` from `rdkit` as inputs.
```

## Full processing steps

```python
import json
import re
import typing

import datasets
import pandas as pd
import rdkit
from rdkit import Chem as rdChem
from tqdm.auto import tqdm


# TODO: Update to 2024.03.6 release when available instead of suppressing warning!
#  See: https://github.com/rdkit/rdkit/issues/7625#
rdkit.rdBase.DisableLog('rdApp.warning')

def count_rings_and_bonds(
    mol: rdChem.Mol
) -> typing.Dict[str, int]:
    """Counts bond and ring (by type)."""
    
    # Counting rings
    ssr = rdChem.GetSymmSSSR(mol)
    ring_count = len(ssr)
    
    ring_sizes = {}
    for ring in ssr:
        ring_size = len(ring)
        if ring_size not in ring_sizes:
            ring_sizes[ring_size] = 0
        ring_sizes[ring_size] += 1
    
    # Counting bond types
    bond_counts = {
        'single': 0,
        'double': 0,
        'triple': 0,
        'aromatic': 0
    }
    
    for bond in mol.GetBonds():
        if bond.GetIsAromatic():
            bond_counts['aromatic'] += 1
        elif bond.GetBondType() == rdChem.BondType.SINGLE:
            bond_counts['single'] += 1
        elif bond.GetBondType() == rdChem.BondType.DOUBLE:
            bond_counts['double'] += 1
        elif bond.GetBondType() == rdChem.BondType.TRIPLE:
            bond_counts['triple'] += 1
    result = {
        'ring_count': ring_count,
    }
    for k, v in ring_sizes.items():
        result[f"R{k}"] = v

    for k, v in bond_counts.items():
        result[f"{k}_bond"] = v
    return result


"""
    Download data and validation indices from:
        "Score-based Generative Modeling of Graphs via the System of Stochastic Differential Equations"
        https://github.com/harryjo97/GDSS
    > wget https://raw.githubusercontent.com/harryjo97/GDSS/master/data/zinc250k.csv
    > wget https://raw.githubusercontent.com/harryjo97/GDSS/master/data/valid_idx_zinc250k.json
"""
df = pd.read_csv('<PATH TO zinc250k.csv>', index_col=0, encoding='utf_8')
feats = []
for i, row in tqdm(df.iterrows(), total=len(df), desc='RDKit feats', leave=False):
    feat = {'smiles': row['smiles']}
    feat['canonical_smiles'] = rdChem.CanonSmiles(feat['smiles'])
    m = rdChem.MolFromSmiles(feat['canonical_smiles'])
    feat.update(count_rings_and_bonds(m))
    feats.append(feat)
df = pd.merge(df, pd.DataFrame.from_records(feats), on='smiles')
df = df.fillna(0)
for col in df.columns:  # recast ring counts as int
    if re.search("^R[0-9]+$", col) is not None:
        df[col] = df[col].astype(int)
# Re-order columns
df = df[
    ['smiles', 'logP', 'qed', 'SAS', 'canonical_smiles',
     'single_bond', 'double_bond', 'triple_bond', 'aromatic_bond',
     'ring_count','R3', 'R4', 'R5', 'R6', 'R7', 'R8', 'R9', 'R10', 'R12', 'R13', 'R14', 'R15', 'R18', 'R24']]


# Read in validation indices
with open('<PATH TO valid_idx_zinc250k.json>', 'r') as f:
    valid_idxs = json.load(f)
df['validation'] = df.index.isin(valid_idxs).astype(int)

# Create HF dataset
dataset = datasets.DatasetDict({
    'train': datasets.Dataset.from_pandas(df[df['validation'] == 0].drop(columns=['validation'])),
    'validation': datasets.Dataset.from_pandas(df[df['validation'] == 1].drop(columns=['validation'])),
})
dataset = dataset.remove_columns('__index_level_0__')
```