README.md

metadata

license: cc-by-sa-4.0
task_categories:
  - tabular-classification
language:
  - en
tags:
  - molecular data
  - few-shot learning
pretty_name: FS-Mol
size_categories:
  - 1M<n<10M
dataset_summary: >-
  FSMol is a dataset curated from ChEMBL27 for small molecule activity
  prediction. It consists of 5,120 distinct assays and includes a total of
  233,786 unique compounds.
citation: ' @article{stanley2021fs, title={FS-Mol: A Few-Shot Learning Dataset of Molecules}, author={Stanley, Matthew and Ramsundar, Bharath and Kearnes, Steven and Riley, Patrick}, journal={NeurIPS 2021 AI for Science Workshop}, year={2021}, url={https://www.microsoft.com/en-us/research/publication/fs-mol-a-few-shot-learning-dataset-of-molecules/} } '
configs:
  - config_name: FSMol
    data_files:
      - split: train
        path: FSMol/train-*
      - split: test
        path: FSMol/test-*
      - split: validation
        path: FSMol/validation-*
dataset_info:
  config_name: FSMol
  features:
    - name: SMILES
      dtype: string
    - name: 'Y'
      dtype: int64
      description: Binary classification (0/1)
    - name: Assay_ID
      dtype: string
    - name: RegressionProperty
      dtype: float64
    - name: LogRegressionProperty
      dtype: float64
    - name: Relation
      dtype: string
  splits:
    - name: train
      num_bytes: 491639392
      num_examples: 5038727
    - name: test
      num_bytes: 5361196
      num_examples: 56220
    - name: validation
      num_bytes: 1818181
      num_examples: 19008
  download_size: 154453519
  dataset_size: 498818769

FS-Mol

FS-Mol is a dataset curated from ChEMBL27 for small molecule activity prediction. It consists of 5,120 distinct assays and includes a total of 233,786 unique compounds. This is a mirror of the Official Github repo where the dataset was uploaded in 2021.

Preprocessing

We utilized the raw data uploaded on Github and performed several preprocessing:

1. Sanitize the molecules using RDKit and MolVS (standardize SMILES format)
2. Formatting (Combine jsonl.gz files to one csv/parquet file)
3. Rename the columns ('Property' to 'Y')
4. Convert the floats in 'Y' column to integers
5. Split the dataset (train, test, validation)

If you would like to try our pre-processing steps, run our script.

Quickstart Usage

Load a dataset in python

Each subset can be loaded into python using the Huggingface datasets library. First, from the command line install the datasets library

$ pip install datasets

then, from within python load the datasets library

>>> import datasets

and load the FSMol datasets, e.g.,

>>> FSMol = datasets.load_dataset("maomlab/FSMol", name = "FSMol")
train-00000-of-00001.parquet: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 152M/152M [00:03<00:00, 39.4MB/s]
test-00000-of-00001.parquet: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1.54M/1.54M [00:00<00:00, 33.3MB/s]
validation-00000-of-00001.parquet: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 517k/517k [00:00<00:00, 52.6MB/s]
Generating train split: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 5038727/5038727 [00:08<00:00, 600413.56 examples/s]
Generating test split: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 56220/56220 [00:00<00:00, 974722.00 examples/s]
Generating validation split: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 19008/19008 [00:00<00:00, 871143.71 examples/s]

and inspecting the loaded dataset

>>> FSMol
DatasetDict({
        train: Dataset({
            features: ['SMILES', 'Y', 'Assay_ID', 'RegressionProperty', 'LogRegressionProperty', 'Relation'],
            num_rows: 5038727
        })
        test: Dataset({
            features: ['SMILES', 'Y', 'Assay_ID', 'RegressionProperty', 'LogRegressionProperty', 'Relation'],
            num_rows: 56220
        })
        validation: Dataset({
            features: ['SMILES', 'Y', 'Assay_ID', 'RegressionProperty', 'LogRegressionProperty', 'Relation'],
            num_rows: 19008
        })

})

Use a dataset to train a model

One way to use the dataset is through the MolFlux package developed by Exscientia. First, from the command line, install MolFlux library with catboost and rdkit support

pip install 'molflux[catboost,rdkit]'

then load, featurize, split, fit, and evaluate the catboost model

import json
from datasets import load_dataset
from molflux.datasets import featurise_dataset
from molflux.features import load_from_dicts as load_representations_from_dicts
from molflux.splits import load_from_dict as load_split_from_dict
from molflux.modelzoo import load_from_dict as load_model_from_dict
from molflux.metrics import load_suite

Split and evaluate the catboost model

split_dataset = load_dataset('maomlab/FSMol', name = 'FSMol')

split_featurised_dataset = featurise_dataset(
  split_dataset,
  column = "SMILES",
  representations = load_representations_from_dicts([{"name": "morgan"}, {"name": "maccs_rdkit"}]))

model = load_model_from_dict({
    "name": "cat_boost_classifier",
    "config": {
        "x_features": ['SMILES::morgan', 'SMILES::maccs_rdkit'],
        "y_features": ['Y']}})

model.train(split_featurised_dataset["train"])
preds = model.predict(split_featurised_dataset["test"])

classification_suite = load_suite("classification")

scores = classification_suite.compute(
references=split_featurised_dataset["test"]['Y'],
predictions=preds["cat_boost_classifier::Y"])

Citation

@article{stanley2021fs, title={FS-Mol: A Few-Shot Learning Dataset of Molecules}, author={Stanley, Matthew and Ramsundar, Bharath and Kearnes, Steven and Riley, Patrick}, journal={NeurIPS 2021 AI for Science Workshop}, year={2021}, url={https://www.microsoft.com/en-us/research/publication/fs-mol-a-few-shot-learning-dataset-of-molecules/