A Bioinformatics Data Platform on Snowflake Using Life-Science Ontologies as Context for AI Agents

Life-science ontologies describe concepts such as genes, phenotypes, diseases, and chemical entities under shared identifiers and hierarchies, and over a quarter of a century they have become a foundation for annotation and analysis in bioinformatics. AI agents now assemble research procedures on their own, choosing and running databases and analysis tools, and how well they work depends on the definitions of meaning they can consult, that is, on the context layer. In enterprise data platforms, this layer is increasingly built as an organization-internal operational ontology, as exemplified by Palantir, whereas reference ontologies in the life sciences are maintained by communities outside any single organization. Using Snowflake as an example, we discuss an architecture composed of semantic views that define the meaning of internal data, Cortex Agent that queries through them, and custom tools and the Model Context Protocol (MCP) that reach external ontologies. The results of concept resolution and mapping obtained from outside are stored by a custom tool in tables within Snowflake for each query, and are joined with internal data under the semantic view and under the same access control. Taking target–disease matching in drug discovery and cohort definition in real-world data as hypothetical use cases, we show, as a conceptual design, that the set of entities to be matched changes with how far the ontology hierarchy is traversed, and that the temporal conditions of a cohort are applied to the same records as the concept set obtained from the ontology.