P03-23
Unraveling Microbiome Complexity: A Knowledge Graph Approach to Functional Interpretation in Drug Discovery
Hirokazu NISHIMURA *1, Taku HIRATA1, Maaly NASAR4, Mark STREER4, Michael HUGHES4, Sachiko FURUYA2, Fumihiko OONO3, Ryuuta SAITOU1
1Discovery Technology Laboratories, Mitsubishi Tanabe Pharma Corporation
2Oncology & Immunology Unit, Mitsubishi Tanabe Pharma Corporation
3Business Development Department, Mitsubishi Tanabe Pharma Corporation
4SciBite Ltd.
( * E-mail: nishimura.hirokazu@ma.mt-pharma.co.jp )
The human microbiome holds significant potential for drug discovery and personalized medicine. Advances in metagenomic analysis have significantly enhanced our ability to detect changes in microbial communities. However, a major technical challenge remains in the functional interpretation of these changes. Understanding how specific microbes and their metabolites influence human physiology is still complex and unresolved.
To address this challenge by clearly understanding the relationships between the microbiome and disease based on the molecular mechanism, we have developed a comprehensive knowledge graph database that links documented microbes and their metabolites to human biological functions. Initially, First, we searched MEDLINE and PMC for literature related to microbes and their metabolites, retrieving approximately 80,000 relevant articles. Utilizing SciBite's named entity recognition (NER) and extraction engine, TERMite, we extracted 10 types of nodes (Microbe, Metabolite, Indication, Gene, Pathway, Gene Ontology, etc.), totaling 538,527 nodes, and mapped 6,852,832 edges. Machine Learning (ML) algorithms and Large Language Models (LLM) were then employed to classify and score these edges.
Our knowledge graph database offers several key advantages. First, it integrates and organizes data from various studies, providing a unified platform for researchers to explore microbial functions and their impacts on human health. Second, the graphical representation of data allows for intuitive visualization of relationships, making it easier to identify potential causal links between microbiome alterations and physiological outcomes. These advantages support hypothesis generation for elucidating disease mechanisms and identifying therapeutic targets.