Specimen, biological structure, and spatial ontologies in support of a Human Reference Atlas

Abstract

The Human Reference Atlas (HRA) is defined as a comprehensive, three-dimensional (3D) atlas of all the cells in the healthy human body. It is compiled by an international team of experts who develop standard terminologies that they link to 3D reference objects, describing anatomical structures. The third HRA release (v1.2) covers spatial reference data and ontology annotations for 26 organs. Experts access the HRA annotations via spreadsheets and view reference object models in 3D editing tools. This paper introduces the Common Coordinate Framework (CCF) Ontology v2.0.1 that interlinks specimen, biological structure, and spatial data, together with the CCF API that makes the HRA programmatically accessible and interoperable with Linked Open Data (LOD). We detail how real-world user needs and experimental data guide CCF Ontology design and implementation, present CCF Ontology classes and properties together with exemplary usage, and report on validation methods. The CCF Ontology graph database and API are used in the HuBMAP portal, HRA Organ Gallery, and other applications that support data queries across multiple, heterogeneous sources.

Fig. 1

From Real-World Entities via Standardized Data to CCF Ontology used by HRA Tools. (a) Human Reference Atlas construction takes real-world data and represents it in standardized data structures that are defined by the interlinked Biological Structure, Spatial, and Specimen data in the CCF Ontology. Anatomical structures in the ASCT+B tables are linked via part_of relationships resulting in a partonomy; they are crosswalked to 3D reference organs in support of spatial queries and exploration using different HRA tools shown on the right. (b) HRA usage includes search for specific cell types or biomarker expression values across organs in a 3D reference space. Users can also upload new experimental data (e.g., a new tissue block from a human donor specimen); the tissue is registered into the HRA by spatially mapping it to a 3D reference organ. If the Registration User Interface (RUI) shown on left is used, anatomical structure tags (see UBERON ID and Label) are automatically assigned based on collision events; spatial search becomes possible in the Exploration User Interface (EUI); and cell types or biomarkers associated with the colliding anatomical structures can be retrieved via ASCT+B tables and explored in the EUI.

Fig. 2

HRA reference organ description. The Visible Human Project female left kidney (#VHFLeftKidney) reference organ is represented as a Spatial Entity class of type Spatial Object Reference, and with a Spatial Placement. Properties such as representation_of with PURL link to UBERON:0004538 (left kidney) refer to Biological Structure data, whereas properties such as organ_owner_sex with PURL link to LNC:LA3–6 (female) refer to Specimen data.

Fig. 3

HRA tissue block description. A Donor (which is_a Person) provides a Sample of type Tissue Block with properties such as Biological Sex and annotation properties such as Creator to keep track of provenance. The Tissue Block is registered with the Biological Structure using annotation property collides_with with three PURL links to UBERON:0002015 (kidney capsule), UBERON:0001225 (cortex of kidney), and UBERON:0002189 (outer cortex of kidney) that all have GLB files with proper Spatial Placement in the 3D Reference Object Library.

Fig. 4

Data specification diagram. Major entities and their relationships are shown as used in JSON-LD. Note that a Sample comes_from a Donor; it is either a Tissue Block or a Tissue Section. A Dataset might be generated from a Sample. A Sample typically has a Spatial Entity which defines its size in relation to a Spatial Object Reference via a Spatial Placement.

Fig. 5

Applications. (a) Registration User Interface (RUI) can be used to assign specimen, biological structure, and spatial metadata. (b) Exploration User Interface (EUI) supports specimen, biological structure, and spatial search and exploration. (c) Mesh-level collision detection improves semantic tagging. (d) HRA Organ Gallery with life-size 3D representations of the Visible Human Project dataset.

Fig. 6

CCF Ontology Generation Pipeline. ASCT+B tables are validated and semantically enriched using the Validation Tool; in parallel, 3D Reference Organ data and crosswalk to ASCT+B tables are processed to compute the HRA. Sample Registration Data tissue blocks with RUI locations and anatomical structure tags can be retrieved for HRA construction or during HRA usage.