ROCKET: Risk-Oriented Causal Knowledge–Enriched Topology for Multi-Omics Digital Twin Modeling

ROCKET-KG is a risk-scored biomedical knowledge graph that integrates 15 curated biological and clinical databases, including Open Targets Platform, CAUSALdb2, DrugBank, EpiGraphDB, GWAS Catalog, and ClinVar.The graph comprises 2,408,551 nodes and 13,341,785 edges, each assigned a continuous ROCKET score in the range [0,1]. This score reflects a composite assessment of five factors: evidence strength, source credibility, topological importance, biological relevance, and functional coherence. ROCKET-KG represents 14 node types (such as genes, diseases, drugs, SNPs, pathways, and biomarkers) and 52 relation types capturing causal, regulatory, and associative interactions across molecular, cellular, and clinical levels. It is designed to enable risk-aware graph reasoning, supporting applications such as disease prediction, patient digital twin modeling, and graph-based retrieval-augmented inference. As a validated subset, ROCKET-CRKG (Causal Risk Knowledge Graph) is a high-quality subgraph containing 2,799,912 edges restricted to four core entity types: gene/protein, risk gene, drug, and disease. Each edge is annotated with two complementary metrics: a causal score representing evidence strength, and the ROCKET score representing overall edge quality. The graph includes 10 biologically meaningful relation types and relies exclusively on standardized identifiers (e.g., DrugBank, Ensembl, NCBI, MONDO, MeSH, UMLS, HPO, GO, and ClinVar). Non-standard or weakly mapped entities have been removed to ensure consistency. ROCKET-CRKG is fully curated with no missing values, duplicate edges, or self-loops, providing a clean, high-confidence dataset for disease risk stratification, drug target discovery, and knowledge graph–driven clinical prediction.