NeDRex: Multi-Domain Knowledge Graph for Drug Repurposing

• NeDRex is a heterogeneous biomedical knowledge graph that integrates genes, diseases, drugs, pathways, tissues, and genomic variants for comprehensive translational research.
• It supports advanced network algorithms such as DIAMOND for module detection, TrustRank for drug-target proximity, and closeness centrality for drug ranking.
• NeDRex enables seamless biomedical reasoning via ChatDRex, offering modular, interoperable, and no-code network analytics for diverse research applications.

NeDRex (Network-based Drug Repurposing and Exploration) is a heterogeneous systems-medicine knowledge graph (KG) designed as the core data resource within the NeDRex platform for supporting network-based disease module detection and drug repurposing workflows. It integrates diverse biomedical entities and relationships to facilitate module-detection algorithms, drug-target proximity analyses, and supports conversational, no-code network analytics within the ChatDRex system. NeDRex serves as the primary backend for complex queries and network algorithms, enabling seamless biomedical reasoning across multiple expert domains (Süwer et al., 26 Nov 2025).

1. Architecture and Design Objectives

NeDRex is architected to unify a broad spectrum of biomedical entities, specifically genes/proteins, diseases, drugs, pathways, tissues, and genomic variants, along with their interactions and ontological hierarchies. The core objectives are:

• Integration of heterogeneous data sources to bring together genes/proteins, diseases, drugs, pathways, tissues, variants, and their multifaceted associations.
• Provision of a single, queryable KG for supporting disease module-detection algorithms (e.g., DIAMOND), network-based drug-target proximity measures (TrustRank, closeness centrality), and downstream drug ranking.
• Reduction of workflow fragmentation by supplying standardized identifiers and controlled vocabularies for all incorporated entities.
• Enabling accessibility and extensibility via a Neo4j backend and the programmatic NeDRexAPI.

A plausible implication is enhanced interoperability across specialized analytic agents, improving collaborative, cross-domain biomedical exploration.

2. Data Sources and Ontological Foundations

NeDRex harmonizes data from several established biomedical repositories and ontologies, constructing composite node and edge sets through curated mappings. The named sources include:

• MyGene.info: Provides gene identifier resolution, enabling mapping across Entrez, Ensembl, and alternate gene nomenclatures.
• UniProt: Supplies protein attributes and cross-references to functional and structural annotations.
• MONDO: Delivers disease nodes with persistent identifiers (exemplified by mondo.0007739 for Huntington’s disease) and hierarchical taxonomy.
• Gene Ontology (GO) and KEGG: Facilitate functional annotation of genes and proteins; used by downstream agents for module validation.
• Pathway databases (via KEGG IDs): Define biological pathway nodes and link relevant proteins.
• Clinical indication and contraindication data: Support drug ↔ disease and drug ↔ phenotype relationships.
• Implicit vocabularies: Tissue annotations, genomic variants (such as SNPs), and side-effect/phenotype nodes.

This consolidation enables multifaceted representation and querying capability for complex translational research scenarios.

3. Graph Schema: Nodes, Edges, and Hierarchies

The NeDRex graph embodies a carefully defined schema, encompassing major node types and semantic relation labels:

Node Type	Relation Label	Description
Gene (protein)	:GeneAssociatedWithDisorder, :Encodes, :InteractsWith	Gene-disease, gene-protein, PPI relationships
Disorder (disease)	:HasIndication, :HasContraindication, :IsSubtypeOf	Disease-drug, disease taxonomy
Drug	:HasIndication, :HasContraindication	Drug-disease clinical links
Pathway	:InPathway	Protein-pathway membership
Tissue	:ExpressedIn	Gene/protein-tissue specificity
GenomicVariant	:Associated	Variants linked to entities
Phenotype/Side-effect	:Associated	Implicit clinical annotations

Hierarchical structure is present, exemplified by disease monophyly via MONDO’s :IsSubtypeOf edges, and the integration of GO/KEGG vocabularies via attribute-linked subgraphs.

A plausible implication is that NeDRex supports inferencing over disease subtypes, pathway memberships, and variant associations within complex translational workflows.

4. Major Subgraphs and Graph Statistics

NeDRex is organized into several “major subgraphs,” which correspond to the principal node-type clusters:

• Genes/Proteins
• Diseases/Disorders
• Drugs
• Pathways
• Tissues
• Genomic Variants

Absolute node and edge counts are not enumerated in (Süwer et al., 26 Nov 2025). For quantitative details, previous publications or direct database introspection are required. The schema, however, is demonstrably capable of supporting large-scale multimodal biomedical network analyses.

5. Supported Network Algorithms

NeDRex enables several network-centric computational methodologies via ChatDRex agents and the NeDRexAPI:

• DIAMOND (Disease Module Detection): Given seed genes associated with a disease, the method iteratively expands the subnetwork by adding nodes with highest connectivity significance to the growing module. The process operationalizes module discovery for phenotype-driven analyses.
• TrustRank (Network-Proximity Ranking): Designed to rank candidate drug targets, TrustRank propagates “trust” from seed disease proteins across the protein-protein interaction (PPI) subgraph. While the precise formalism is not detailed in (Süwer et al., 26 Nov 2025), the algorithm is a personalized PageRank variant.
• Closeness Centrality (Proximity Measure): For each candidate drug node $d$, the closeness is computed by

$$C_{\mathrm{closeness}(d)} = \frac{N-1}{\sum_{v\ne d} \mathrm{dist}(d,v)}$$

where $N$ is the number of nodes, and $\mathrm{dist}(d, v)$ denotes the shortest-path distance. Drugs are ranked by inverse average distance to disease module proteins.

A plausible implication is that network-based ranking provides robust, interpretable candidate discovery for drug repurposing tasks.

6. Integration with Conversational Multi-Agent Systems

NeDRex is the knowledge backbone for ChatDRex, a conversation-based multi-agent environment. Integration features include:

• Query Routing: The Planning agent discerns user intent, delegating KG-centric questions to the NeDRex agent.
• Data Retrieval: The NeDRex agent translates queries into “question graphs” (node sets and filter flags), utilizes embedding-based lookups for string similarity, and executes deterministic Cypher queries against the Neo4j store.
• Algorithm Execution: For analytics (DIAMOND, TrustRank, Closeness), the NeDRex agent uses API endpoints, passing resolved gene identifiers.
• Result Finalization: The Finalize agent formats outputs as Markdown or interactive visualizations, applies hallucination guardrails, and streams results to users.

This design ensures modularity and transparency in biomedical graph querying and analysis, supporting both human-in-the-loop workflows and automated validation.

7. Representative Applications and Case Studies

NeDRex underpins several disease-focused drug repurposing workflows as demonstrated in (Süwer et al., 26 Nov 2025):

• Huntington’s Disease Workflow:
  • Gene discovery via Cypher queries
  • Module identification with DIAMOND
  • Drug ranking with TrustRank and/or Closeness
• Alzheimer’s Disease Demonstration:
  • Seed genes selected (APOE, APP, etc.)
  • Module expansion via DIAMOND
  • Functional coherence assessment using DIGEST (leveraging GO/KEGG annotations in KG)
  • Drug prioritization followed by literature retrieval

These examples illustrate NeDRex’s utility as a persistent and semantically rich knowledge graph, enabling hypothesis generation, module and proximity analysis, and supporting translational research via accessible, no-code interaction (Süwer et al., 26 Nov 2025).