About B3DB

In this repo, we present a large benchmark dataset, Blood-Brain Barrier Database (B3DB), compiled from 50 published resources (as summarized at raw_data/raw_data_summary.tsv) and categorized based on the consistency between different experimental references/measurements. This dataset was published in Scientific Data and this repository is occasionally uploaded with new experimental data. Scientists who would like to contribute data should contact the database's maintainers (e.g., by creating a new Issue in this database).

A subset of the molecules in B3DB has numerical logBB values (1058 compounds), while the whole dataset has categorical (BBB+ or BBB-) BBB permeability labels (7807 compounds prior to v1.0.0 and 7982 compounds after). Some physicochemical properties of the molecules are also provided.

Citation

Please use the following citations in any publication using our B3DB dataset:

@article{Meng_A_curated_diverse_2021,
author = {Meng, Fanwang and Xi, Yang and Huang, Jinfeng and Ayers, Paul W.},
doi = {10.1038/s41597-021-01069-5},
journal = {Scientific Data},
number = {289},
title = {A curated diverse molecular database of blood-brain barrier permeability with chemical descriptors},
volume = {8},
year = {2021},
url = {https://www.nature.com/articles/s41597-021-01069-5},
publisher = {Springer Nature}
}

@article{Meng_B3clf_2025,
author = {Meng, Fanwang and Chen, Jitian and Collins-Ramirez, Juan Samuel and Ayers, Paul W.},
doi = {10.26434/chemrxiv-2025-xschc},
journal = {ChemRxiv},
number = {to be updated pending peer-reviewed publication},
title = {B3clf: A Resampling-Integrated Machine Learning Framework to Predict Blood-Brain Barrier Permeability},
volume = {to be updated pending peer-reviewed publication},
year = {to be updated pending peer-reviewed publication},
url = {to be updated pending peer-reviewed publication},
publisher = {to be updated pending peer-reviewed publication}
}

Features of B3DB

1. The largest dataset with numerical and categorical values for Blood-Brain Barrier small molecules (to the best of our knowledge, as of February 25, 2021).

2. Inclusion of stereochemistry information with isomeric SMILES with chiral specifications if available. Otherwise, canonical SMILES are used.

3. Characterization of uncertainty of experimental measurements by grouping the collected molecular data records.

4. Extended datasets for numerical and categorical data with precomputed physicochemical properties using mordred.

Usage

Via PyPI

The B3DB dataset is avaliable at PyPI. One can install it using pip:

pip install qc-B3DB

Then load the data (dictionary of pandas dataframe) with the following code snippet:

from B3DB import B3DB_DATA_DICT

# access the data via dictionary keys
# 'B3DB_regression'
# 'B3DB_regression_extended'
# 'B3DB_classification'
# 'B3DB_classification_extended'
# "B3DB_classification_external"
df_b3db_reg = B3DB_DATA_DICT["B3DB_regression"]
df_b3db_reg.head()
#    NO.                                      compound_name  ... group comments
# 0    1                                         moxalactam  ...     A      NaN
# 1    2                                      schembl614298  ...     A      NaN
# 2    3                             morphine-6-glucuronide  ...     A      NaN
# 3    4  2-[4-(5-bromo-3-methylpyridin-2-yl)butylamino]...  ...     A      NaN
# 4    5                                                NaN  ...     A      NaN

# [5 rows x 10 columns]

Manually Download the Data

There are two types of dataset in B3DB, regression data and classification data and they can be loaded simply using pandas. For example

import pandas as pd

# load regression dataset
regression_data = pd.read_csv("B3DB/B3DB_regression.tsv",
                              sep="\t")

# load classification dataset
classification_data = pd.read_csv("B3DB/B3DB_classification.tsv",
                                  sep="\t")

# load extended regression dataset
regression_data_extended = pd.read_csv("B3DB/B3DB_regression_extended.tsv.gz",
                                       sep="\t", compression="gzip")

# load extended classification dataset
classification_data_extended = pd.read_csv("B3DB/B3DB_classification_extended.tsv.gz",
                                           sep="\t", compression="gzip")

Examples in Jupyter Notebooks

We also have three examples to show how to use our dataset, numerical_data_analysis.ipynb, PCA_projection_fingerprint.ipynb and PCA_projection_descriptors.ipynb. PCA_projection_descriptors.ipynb uses precomputed chemical descriptors for visualization of chemical space of B3DB, and can be used directly using MyBinder, . Due to the difficulty of installing RDKit in MyBinder, only PCA_projection_descriptors. ipynb is set up in MyBinder.

Data Curation

Detailed procedures for data curation can be found in data curation section in this repository.

The materials and data under this repo are distributed under the CC0 Licence.

ChangeLog

• 2025Aug16, the B3DB dataset is avaliable via PyPI.
• 2025Aug16, we have added a new set of 171 BBB+ and 4 BBB- compounds to the dataset since version 1.1.0.