Hyperparameter Settings for a Biomedical Knowledge Graph Drug Discovery Model

245,235 entities and 7,155,373 triples form a cross-medicine knowledge graph for Type 2 Diabetes Mellitus research. Zekun Zhou published this 5.5 KB Excel file in May 2026, containing hyperparameter settings used to evaluate graph embedding models like ComplEx for link prediction. The work integrates data from Hetionet, SymMap, TCMBank, STRING, and TTD.

Use Cases

• Reproducing the hyperparameter sensitivity analysis for the ComplEx graph embedding model mentioned in the description
• Benchmarking other graph embedding models against the reported MRR and Hits@K performance metrics
• Investigating the unified path scoring framework incorporating length decay, node weights, and experimental evidence bonuses
• Analyzing the impact of the Ingredient Specificity Index (ISI) and hybrid path confidence calibration on mechanistic path discovery

Strengths

• The underlying knowledge graph is large, containing 15 entity types and 245,235 entities
• Model performance is quantified with specific metrics (MRR = 0.213 ± 0.004, Hits@10 = 0.418 ± 0.003)
• The dataset has a clear, permissive license (CC-BY-4.0)

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• The 5.5 KB file size suggests a very limited scope, likely containing only configuration parameters, not the full knowledge graph data

Provenance

Source

Zekun Zhou via figshare

Collection Method

Parameters derived from constructing and evaluating a cross-medicine knowledge graph integrating multiple biomedical sources.

Freshness

Last updated 2026-05-13 17:33:23; freshness should be verified