From Static Biomedical Knowledge Graphs to Dynamic Queries with ODE Models

Session track: Modern Applications

Session time: 09:30 pm - 10:00 pm UTC November 12

Session description:

Biomedical knowledge graphs are often queried as static structures, even when the biological processes they describe are dynamic. In this session, Thomas Sauer will describe an implementation pattern for using ordinary differential equation models to guide traversal through biomedical knowledge graphs in Neo4j. The session will show how graph relationships and model state can play complementary roles. A knowledge graph can encode relationships among entities such as proteins, anatomical structures, biological processes, phenotypes, and interventions. An ODE model can then provide time-dependent context: which quantities are increasing, which are decaying, and which model states are most relevant at a particular point in a simulation. Thomas will explain how those simulated states can be mapped onto graph entities and used to prioritize, constrain, or interpret Cypher traversals. You will learn a practical approach for connecting equation-based models to Neo4j knowledge graphs, including how to define the boundary between the model and the graph, how to represent model variables as graph-linked concepts, how to use model outputs as query inputs, and how to evaluate whether the resulting traversals are scientifically meaningful. The session will also discuss limitations encountered when connecting dynamic models to static graph data, including incomplete mappings, ambiguous biological relationships, and the difference between graph connectivity and mechanistic evidence.