Papyrus: 60 Million Small Molecule Bioactivity Datapoints for Predictive Modeling

60 million datapoints comprise the Papyrus dataset, an aggregated collection of small molecule bioactivities for machine learning. It combines large public sources like ChEMBL and ExCAPE-DB with smaller high-quality datasets, standardized for predictive modeling. The dataset was created by Olivier J. M. Béquignon of Leiden University.

Use Cases

• Train quantitative structure-activity relationship (QSAR) models based on aggregated bioactivity data.
• Perform proteochemometric modeling based on combined ligand and protein information.
• Construct benchmark predictive models based on the pruned and standardized data collection.
• Filter bioactivity data for specific research questions based on the described variety of filtering methods.

Strengths

• Approximately 60 million datapoints provide a large-scale resource.
• Aggregates multiple public datasets including ChEMBL and ExCAPE-DB, increasing coverage.
• Data has been standardized and normalized in a manner suitable for machine learning.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Aggregated from public datasets including ChEMBL and ExCAPE-DB.

Collection Method

Data aggregation, standardization, and normalization.