BBB Permeability · maomlab
B3DB
Curated blood-brain barrier permeability measurements and chemical descriptors for central nervous system drug discovery.
What B3DB is
The Blood-Brain Barrier Database (B3DB) compiles experimental blood-brain barrier (BBB) permeability measurements collected from 50 published literature sources. Developed to support central nervous system (CNS) drug discovery and computational screening, it provides standardised chemical identifiers alongside permeability labels and physicochemical descriptors.
The dataset contains 7,807 small molecules annotated with categorical permeability labels (BBB+ or BBB-). A subset of 1,058 compounds also includes numerical logBB values, representing the logarithmic ratio of drug concentrations between the brain and blood. Data records are grouped to capture measurement consistency across differing literature references.
Originally published by Fanwang Meng and colleagues and maintained in the theochem/B3DB repository, this version distributed by maomlab mirrors the 2021 classification and 2023 regression releases while partitioning entries using a realistic splitting strategy.
How it was made
Experimental values and chemical structures were gathered from 50 independent publications. Molecular structures are documented with InChI strings, IUPAC names, compound names, and SMILES representations. Stereochemistry is preserved using isomeric SMILES with chiral specifications when reported; otherwise, canonical SMILES are provided.
Records are stratified into groups according to measurement agreement across sources to capture experimental uncertainty. Extended datasets include molecular descriptors precalculated with the Mordred package. Data splits into training and testing sets (comprising 5,855 train and 1,951 test examples for classification) follow the realistic split methodology proposed by Martin et al. (2018).
Key facts
- Total molecules
- 7,807
- Numerical logBB entries
- 1,058
- Data format
- Tabular dataset with SMILES, InChI, and metadata
- Task categories
- Tabular classification and tabular regression
- Data sources
- 50 published scientific resources
- Descriptor engine
- Mordred
- Splitting strategy
- Realistic Split (Martin et al., 2018)
- Repository publisher
- maomlab
What you can use it for
- Categorical Permeability Screening
- Train binary classification models to predict whether candidate drug molecules cross the blood-brain barrier (BBB+ versus BBB-).
- Quantitative logBB Prediction
- Build and evaluate regression models on the 1,058-molecule subset with experimental brain-to-blood partition coefficients (logBB).
- Chemoinformatics Benchmarking
- Benchmark machine learning algorithms, molecular fingerprints (such as Morgan or MACCS keys), and physicochemical descriptors for CNS compound filtering.
Explore the data
The schema, real rows, full-text search and a SQL filter — read in your browser, nothing downloaded.
Explore
Read the real rows without downloading anything
Reading rows…
LiveRead from maomlab/B3DB at the moment you asked. Nothing is cached or stored — every row above came from that request.
Strengths and limits
Good at
- Provides both categorical binary labels across 7,807 molecules and numerical logBB measurements for 1,058 compounds.
- Preserves stereochemical details by applying isomeric SMILES with chiral annotations whenever documented in literature.
- Characterises uncertainty across differing experimental reports by grouping records based on cross-reference consistency.
- Precomputed physicochemical descriptors generated via the Mordred package are included for modelling.
Watch out for
- Numerical logBB measurements are limited to 1,058 of the 7,807 compounds, with the remainder having only binary labels.
- Compiling experimental records across 50 distinct resources introduces variable experimental conditions and measurement protocols across sources.
Questions
Can I look inside B3DB without downloading it?
Yes. This page reads its schema, real rows and per-column statistics in the browser, with full-text search and a SQL filter. Nothing is downloaded to your machine.
What licence is B3DB under?
Its authors declare cc0-1.0. Read the terms on the original page before reusing the data.
What is the difference between the classification and regression subsets?
The full dataset of 7,807 compounds contains binary classification labels denoting whether a compound is permeable (BBB+) or non-permeable (BBB-). A smaller subset of 1,058 compounds provides numerical logBB values for quantitative regression.
What molecular features and identifiers are included in the dataset?
The dataset contains compound names, IUPAC names, SMILES strings, PubChem CIDs, InChI strings, categorical permeability labels, numerical logBB values, reference attribution, cluster assignments, and precomputed Mordred descriptors.
How are the training and test sets partitioned?
The splits were generated using the realistic split approach outlined by Martin et al. (2018), partitioning the classification subset into 5,855 training entries and 1,951 test entries.
How to cite it
Meng, Fanwang and Xi, Yang and Huang, Jinfeng and Ayers, Paul W., 'A curated diverse molecular database of blood-brain barrier permeability with chemical descriptors', Scientific Data, vol. 8, no. 289, 2021. https://doi.org/10.1038/s41597-021-01069-5
Written from B3DB's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.