KG-Orchestra: An Open-Source Multi-Agent Framework for Evidence-Based Biomedical Knowledge Graphs Enrichment

1. Abstract Biomedical Knowledge Graphs (BKGs) offer integrative representations of complex biology, yet their utility is compromised by the limitations of current construction methods: manual curation offers high fidelity but is unscalable, whereas purely automated Large Language Model (LLM) approaches often yield broad networks lacking mechanistic granularity. We present KG-Orchestra, an open-source multi-agent framework designed to build specialized, directional, cause-and-effect BKGs by enriching seed graphs. The framework focuses on increasing granularity within specific topics by leveraging Retrieval-Augmented Generation (RAG) to autonomously acquire, validate, and integrate evidence. The system orchestrates specialized agents for retrieval, schema alignment, and triplet validation with explicit, traceable provenance, transforming sparse seeds into dense, high-resolution resources. We evaluated KG-Orchestra on two specialized contexts—the mechanistic link between Nelivaptan and Alzheimer’s Disease (NADKG) and the complex probiotic interactions within the gut–brain axis (ProPreSyn-GBA)—across varying computational budgets. Our benchmarking results demonstrate that Qwen 3 variants deliver superior reasoning performance and that hybrid retrieval strategies significantly enhance evidence relevance. Furthermore, the multi-agent architecture ensures high triplet integrity and biological validity through iterative cross-checking and self-correction. The framework remains computationally flexible, deploying from single laptop GPUs to high-performance clusters. By bridging knowledge gaps and adding context-aware entities, KG-Orchestra increases reliability while validating seed assertions against up-to-date sources. This versatility supports critical downstream applications, including completing missing mechanistic pathways, integrating novel entities for drug repurposing, constructing targeted subgraphs from entity lists, and retroactively validating graph evidence for transparent auditing. Competing Interest Statement The authors have declared no competing interest. 11. Data Availability The data related to this article are available as an open-source Python package at https://github.com/Fraunhofer-SCAI-Applied-Semantics/KG-Orchestra under the Apache 2.0 License. 13. Glossary - Ablation Study - A scientific procedure used to determine the contribution of individual components of an AI system by removing them one at a time and measuring the resulting impact on performance. - AI Agent - An autonomous computational entity driven by a Large Language Model that is designed to perform a specific, specialized task—such as text retrieval, schema alignment, or fact-checking—within a larger workflow. - Biomedical Knowledge Graph (BKG) - A structured network representation of biological knowledge where nodes represent entities (e.g., genes, drugs, diseases) and edges represent the semantic or causal relationships between them. - Embedding (Dense vs. Sparse) - A mathematical representation of text as a numerical vector. Dense embeddings capture semantic meaning and context, while sparse embeddings (like SPLADE) focus on specific keyword matching and lexical importance. - Evidence - Within this framework, a specific paragraph-level textual excerpt retrieved from scientific literature that provides direct support for, or contradiction of, a proposed knowledge assertion. - Hybrid Retrieval - A search methodology that combines both dense and sparse embeddings to improve the accuracy and relevance of retrieved documents by balancing semantic intent with keyword precision. - Knowledge Graph Enrichment - The process of expanding an existing knowledge graph by adding new entities, discovering novel relationships, or validating existing assertions using external data sources. - Large Language Model (LLM) - A type of artificial intelligence trained on massive datasets of text to understand, generate, and reason with natural language (e.g., Qwen, GPT-4, DeepSeek). - Multi-Agent System (MAS) - An architectural framework where multiple specialized AI agents collaborate to solve complex tasks by decomposing them into smaller, manageable sub-processes. - NDCG (Normalized Discounted Cumulative Gain) - A standard metric used in information retrieval to measure the quality of ranking; it rewards systems for placing the most relevant results at the top of a list. - Ontology - A formal and standardized set of definitions, categories, and properties that dictate how entities and relationships should be structured within a specific domain. - Provenance - The documented history of a data point, including its original source (e.g., DOI or PubMed ID) and the computational steps taken to extract and validate it. - Retrieval-Augmented Generation (RAG) - A technique that enhances LLM performance by retrieving relevant documents from an external corpus and providing them to the model as context before generating a response. - Schema Alignment - The process of mapping newly extracted entities and relationships to the predefined structure and vocabulary of an existing knowledge graph to ensure consistency. - Seed Knowledge Graph - An initial, often smaller, knowledge graph that serves as the foundation and structural guide for further enrichment and expansion. - Triplet - The fundamental unit of a knowledge graph, consisting of three parts: a Head (subject), a Relation (predicate), and a Tail (object), forming a statement such as (Drug A — treats — Disease B).