Semantic Entity Mapping

Connect disparate semantic spaces by aligning the same real-world entities across different identifier schemes, data models, and ontologies—so your data becomes interoperable end-to-end.

The Problem

Your data uses one set of identifiers. Your collaborators use another. Public databases use a third. Without mappings between them, integration is manual, error-prone, and doesn’t scale.

Semantic entity mapping solves this by systematically aligning entities across identifier schemes, data models, and ontologies—so “the same thing” is recognized as such, wherever it appears.

What You Get

A complete, validated mapping set connecting your source entities to your target vocabularies, delivered in SSSOM format with full provenance. Depending on your needs, this can be:

  • A one-off mapping for a specific integration project
  • A continuous mapping system that stays current as source and target resources evolve

How It Works

1. Scope Definition

I start by agreeing on boundaries with you:

  • What’s being mapped? (e.g., all disease concepts, a subset of assays, specific entity types)
  • Where are they mapping to? (e.g., a particular ontology branch, multiple target vocabularies)
  • What precision is required? Do you need exact semantic matches, or is controlled conflation acceptable for your use case?

2. Automated Candidate Generation

I generate candidate mappings using multiple approaches:

  • Lexical matching (string similarity, normalization)
  • Semantic similarity (embeddings, structural features)
  • Mapping hops via existing mapping sets
  • Hybrid methods combining multiple signals

Every candidate is captured with rich metadata—match confidence, method used, supporting evidence—using the SSSOM standard I helped develop.

3. Iterative Refinement

Raw candidates aren’t final mappings. I refine them through:

  • Constraint enforcement: one target per source, or multiple where appropriate
  • Agentic workflows: Automated filtering using evidence and provenance to surface high-confidence matches and flag ambiguous cases
  • Human review: Expert judgment on mappings that require domain knowledge

The level of human oversight is adjustable based on your quality requirements.

4. Delivery

You receive:

  • Validated mapping files in SSSOM format
  • Documentation of scope decisions and methodology
  • Quality metrics and coverage reports
  • Training for your team (if maintaining mappings in-house)
  • Pipeline integration (optional)

For continuous mappings, I establish update cycles and change detection so your mappings evolve with your data.

Why This Approach

Most mapping efforts fail in one of two ways: they’re too automated (generating garbage), or too manual (never finishing). This methodology balances both—using automation to scale, and human expertise where it matters.

I led the development of the SSSOM standard precisely because I kept seeing the same problem: teams would spend months producing mappings, then lose the rationale behind every decision. Provenance isn’t a nice-to-have—it’s what separates a mapping set you can maintain from one you’ll have to redo next year.

The result: mappings you can trust, with clear provenance for every decision.

Case Study

See how Mondo maintains 30,000+ mappings to external disease resources through automated matching and structured human review in this case study.

Get in touch to discuss your mapping needs.