MSAT: A Heterogeneous Graph for Predicting Chinese Materia Medica Adverse Reactions

A heterogeneous graph dataset integrating Chinese Materia Medica, compounds, protein targets, and adverse drug reactions for pharmacovigilance prediction. The dataset, authored by Bowen Shi and last updated in March 2026, contains 27,062 curated CMM-ADR associations used to train and evaluate the MSAT graph neural network model.

Use Cases

• Train heterogeneous graph neural networks for adverse drug reaction prediction based on integrated CMM-compound-target-ADR relationships.
• Benchmark pharmacovigilance models under class imbalance based on the described 1:10 positive-to-negative ratio stress test.
• Generate hypotheses for post-marketing safety surveillance of Chinese Materia Medica based on predicted high-confidence ADR signals.
• Study the integration of statistical evidence (FAERS report counts) with biological mechanisms (protein targets) in a single graph framework.

Strengths

• Contains 27,062 curated Chinese Materia Medica–adverse drug reaction associations.
• Edges are augmented with a six-dimensional evidence feature vector, including FAERS report counts and semantic similarity.
• Model trained on this data achieved a high AUC of 0.9792 in stratified 10-fold cross-validation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the underlying graph nodes and edges is unknown, which may limit suitability assessment.
• The primary file is a 383.9 KB PDF, suggesting the dataset itself may be described rather than fully provided.

Provenance

Source

figshare

Collection Method

Constructed from an evidence-rich heterogeneous graph integrating CMMs, compounds, protein targets, and ADRs, with FAERS data.

Time Range

The dataset description references a model evaluation performed in 2026, but the underlying FAERS data time range is unspecified.

Freshness

Last updated 2026-03-18 07:45:21; freshness should be verified.

Geography

Focus is on Chinese Materia Medica, but the pharmacovigilance evidence (FAERS) is from the U.S. FDA.