31 ChEMBL Compound Activity Data Sets for Regression Modeling

Jenny Balfer at the University of Bonn curated 31 data sets from ChEMBL version 17 for regression modeling. The data contains compounds active against human targets in direct inhibition/binding assays with high confidence scores and Ki values below 100 micromolar. Duplicates, pan-assay interference compounds, and other reactive molecules were removed, and only sets with at least 500 compounds were included.

Use Cases

• Train regression models to predict compound binding affinity (Ki values) based on chemical structure.
• Benchmark machine learning algorithms for bioactivity prediction using curated, high-confidence assay data.
• Develop quantitative structure-activity relationship (QSAR) models for specific human protein targets.
• Analyze the distribution of active compounds across different biological targets for hit identification.

Strengths

• Data is curated from ChEMBL version 17, a well-known bioactivity database.
• Includes only high-confidence data from direct inhibition/binding assays against human targets.
• Applied quality filters: removed duplicates, pan-assay interference compounds, and reactive molecules.
• Each included data set contains at least 500 compounds, providing a minimum scale for modeling.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific modeling needs.
• Last update date is unknown; freshness unverified.

Provenance

Source

University of Bonn, derived from ChEMBL version 17.

Collection Method

Selected and filtered from the ChEMBL database based on specific activity and confidence criteria.

Time Range

ChEMBL version 17 suggests a snapshot from a specific historical release; exact temporal coverage of the underlying data is null.

Freshness

Last updated date is unknown.

Geography

null