ApiNATOMY: Topological Modeling for Physiology

• ApiNATOMY is a framework for representing multiscale physiological circuit maps using typed graphs and layered anatomical conduits.
• It combines a formal knowledge representation model with web-based knowledge management tools to convert high-level anatomical sketches into detailed, machine-readable models.
• The system integrates with external ontologies and knowledge graphs, supporting semantic web standards and FAIR data practices across biological systems.

ApiNATOMY is a framework for the topological and semantic representation of multiscale physiological circuit maps, created to support researchers in mapping data related to the peripheral nervous system and other physiological systems, with an emphasis on their relevance to the organs under investigation. It integrates a Knowledge Representation (KR) model and a suite of Knowledge Management (KM) tools. The KR model enables physiology experts to easily capture interactions between anatomical entities, while the KM tools help modelers convert high-level abstractions into detailed models of physiological processes, which can be integrated with external ontologies and knowledge graphs (Kokash et al., 19 Sep 2025).

1. Scope and architectural organization

ApiNATOMY is organized into two cooperating layers. The first is a Knowledge Representation model, whose core is a labeled, typed graph capturing anatomical conduits (“lyphs”), their internal materials, and the points at which they branch or exchange contents. The second is a suite of Knowledge Management tools, which provide template-based model specification, visualization, validation, and serialization services (Kokash et al., 19 Sep 2025).

The framework is presented for anatomical connectivity, particularly in the context of the nervous system, described as a complex network of nerves and ganglia that plays a critical role in coordinating and transmitting signals throughout the body. The stated emphasis on the peripheral nervous system and on organ relevance situates ApiNATOMY within physiological circuit mapping rather than purely morphological description.

A plausible implication is that the framework is intended to mediate between expert anatomical knowledge, computational representation, and downstream knowledge graph integration. This interpretation is consistent with the explicit coupling of KR and KM layers and with the end-to-end workflow from spreadsheets to RDF/OWL and Neo4j.

2. Formal knowledge representation model

At its heart, an ApiNATOMY model is a directed graph

$G \;=\; (V,\,E)\,,$

where $V$ is the set of nodes, each representing a “point” in the tissue architecture (bifurcation, border, source, target, internal compartment, etc.), and $E\subseteq V\times V$ is the set of links, each of which “conveys” an anatomical conduit. Each link $e\in E$ carries an associated lyph definition $\ell(e)$, where a lyph is a layered compartment made up of materials or other lyphs. If $L$ is the set of all lyph types, the mapping

$\mathit{conveys}: E \;\longrightarrow\; L$

assigns each link its tissue-type conduit. Conversely, each lyph $\lambda\in L$ is connected to its host materials $M$ by

$$\mathit{materialsIn}:\;L\;\longrightarrow\;\mathcal{P}(M)\,,$$

a set-valued “has-material-layer” relation (Kokash et al., 19 Sep 2025).

The principal modeling classes include node classes, link classes, lyph classes, material classes, and chain templates. Node classes include source/target, root/leaf, and internalNodes. Link classes convey conduits, represent lyph borders, or serve purely topological roles such as scaffold wires. Lyph classes are layered tissue compartments, possibly nested, each with layers and internal lyphs. Material classes are chemical substances or composite tissues used to build lyphs. Chain templates are parameterized patterns, including linear vascular or neural chains, that expand to subgraphs.

Relationships are stored as bidirectional properties in the JSON schema. The key examples are

$V$0

Chains introduce a higher-order construct. A chain $V$1 is either specified by an ordered node sequence

$V$2

or by a lyph template plus a level count, which the KM tools automatically unroll into concrete nodes and links. This formalism makes chain templates central to ApiNATOMY’s multiscale representation strategy: high-level patterns can be specified compactly and then expanded into explicit graph structure.

3. Knowledge management tools and workflow

ApiNATOMY’s KM layer is a web-based, JavaScript/React application that guides users from physiologists drafting Gray-style sketches to fully validated, machine-readable anatomical circuits. Model specification begins with Spreadsheet Templates: users start with a multi-sheet XLSX file, and each sheet corresponds to one resource type, namely nodes, links, lyphs, materials, chains, and coalescences. The KM tool parses these sheets into JSON, validates against the ApiNATOMY JSON Schema, and reports any structural errors (Kokash et al., 19 Sep 2025).

Visual modeling and editing are provided by several coordinated components. A Graph Viewer presents $V$3 on a WebGL force-directed 2D/3D canvas, where users can toggle groups, highlight resources, and inspect bidirectional relationships in an auxiliary graph helper. Resource Editors include a Code Editor for direct JSON editing, a Material Editor with a DAG view of material composition and CRUD operations plus ontology annotation, a Lyph Editor with a three-pane tree for lyph hierarchies, layer composition, and internal lyphs with undo/redo and visual diff, and Chain & Coalescence Editors for defining and manipulating chain templates, including chain splitting, merging, lateralization, and overlapping layer definitions. A Scaffolds & Layout Editor allows loading of a scaffold such as the TOO map, which defines anchor points, wires, and regions; connectivity elements can be “anchoredTo” scaffold features to impose anatomical positioning.

Automatic expansion and validation form the next stage of the workflow. Chains and lyph templates are programmatically unrolled into the full subgraph. A rules engine checks for missing identifiers, inconsistent references, and schema violations, and issues warnings or errors in a panel. Serialization and export complete the process: the validated, expanded model is serialized to JSON-LD, including a context that maps ApiNATOMY properties to RDF/OWL IRIs, and a backend pipeline converts JSON-LD to RDF triples and loads them into SciGraph (Neo4j + Blazegraph) as part of the SPARC Knowledge Graph (SCKAN).

This workflow suggests a deliberate progression from high-level abstractions to machine-readable and queryable graph representations. The use of templates, schema enforcement, and serialization services indicates that reproducibility and structural consistency are treated as first-order requirements rather than post hoc additions.

4. Multiscale physiological circuit modeling

A detailed example is the rat bladder-autonomic pathway. Experts begin with three scale sketches: organ-level, spinal cord foldout, and single neuron morphology. The spreadsheet definition includes nodes for ganglia and spinal levels, links for axon segments, and lyphs for nerve sheaths and tissues. A chain template, “Bladder autonomic chain,” is defined by root (Cystic Plexus) → leaf (Pelvic Nerve) via a 5-segment lyph template. The chain’s root is anchored to the vesical region on the TOO map; subsequent nodes follow the blue CSF conduit, then red vascular loop, mimicking topological route. The viewer output is a 3D WebGL rendering where the neuronal chain coils along the scaffold wire, with lyph layers—axon, myelin, and endoneurium—rendered in cross-section (Kokash et al., 19 Sep 2025).

A second example is peripheral nerve topology in the “Neurulator” setting. Upon loading, the “Neurulator” algorithm finds closed subgraphs, such as a dorsal root ganglion plus its afferent and efferent branches, and auto-generates groups. In this representation, chains represent union of all axons in a neuron population, while collaterals and branching are abstracted as multi-hosted nodes. Final models drive the Flatmap pipeline: spectral overlays on Google-Maps style 2D canvases (AC & FC flatmaps) where each nerve route is a colored polyline annotated with Uberon/FMA IDs.

These examples clarify a recurrent point of interpretation. ApiNATOMY does not equate chain instances with single-cell morphology in all cases; in the peripheral nerve topology example, chains represent populations rather than individual cells. The abstraction is explicit, and the corresponding trade-off is discussed elsewhere in the framework as a limitation rather than an unqualified equivalence to per-cell anatomical reconstruction.

5. Ontology and knowledge graph integration

ApiNATOMY is designed to mesh with external ontologies and knowledge graphs through a sequence of data transformations. JSON is converted to JSON-LD via an ApiNATOMY context file that maps each property, such as “conveys” or “hostedBy,” to a persistent IRI. Standard JSON-LD processors then produce a TBox/ABox in RDF, enabling SWRL or OWL DL reasoning. RDF triples are bulk-loaded into Blazegraph or Neo4j via SciGraph’s ETL pipelines (Kokash et al., 19 Sep 2025).

In-tool annotation uses the SciGraph REST API. Editors expose an “ontologyTerms” lookup field that issues Cypher queries such as

$V$4

to retrieve candidate Uberon, FMA, or NIFSTD classes. External definitions can be incorporated through an imports statement of the form $E\subseteq V\times V$6 which pulls in external definitions without duplication and namespaces them by IRI prefixes.

The paper also outlines mapping algorithms and formal rules. Schema-governed converters are described for XLSX→JSON and JSON→Neo4j CSV. Automated chain expansion rules specify that if a chain template has $V$5 levels and a lyph template $V$6, the system generates nodes $V$7 and links $V$8, each conveying a fresh instance of $V$9. Layout constraints are also given: for a scaffold wire $E\subseteq V\times V$0 with anchor endpoints $E\subseteq V\times V$1, a “wired” link $E\subseteq V\times V$2 stretches so that its source and target coincide with $E\subseteq V\times V$3 and $E\subseteq V\times V$4, while intermediate hosted nodes are placed at

$E\subseteq V\times V$5

This architecture places ApiNATOMY at the interface between anatomical modeling and semantic web infrastructure. The explicit JSON-LD context, RDF/OWL serialization, and SciGraph ingestion collectively support interoperability with existing biomedical ontologies and knowledge graphs.

6. Applications, limitations, and prospective extensions

The stated applications include the SPARC Connectivity KB (SCKAN), where ApiNATOMY organizes ANS–organ circuit knowledge for neuromodulation planning and data annotation; Flatmap Generation, where it drives interactive AC/FC flatmaps across multiple species—rat, mouse, pig, and human; Multi-scale Simulation, where, as in de Bono et al. (2017), it supports bond-graph composition of physiological flows along ApiNATOMY circuits; and Educational Modules, where the scaffold and viewer may be repurposed for anatomy teaching (Kokash et al., 19 Sep 2025).

The framework’s limitations are described explicitly. Visual conventions in force-directed, scaffold-constrained layouts differ from traditional anatomical schematics, and some domain experts require training. There is also an abstraction-versus-detail trade-off: collapsing neuron populations into single chains sacrifices per-cell branching detail. Performance is another limitation, since extremely large connectivity graphs, including whole-body and multi-organ cases, can stress WebGL rendering and Neo4j query performance.

Future work is identified in four directions. Quantitative Multi-scale Analysis would integrate bond-graph and finite-element methods for fluid and electrical flows along ApiNATOMY conduits. AI-Assisted Curation would incorporate NLP extracts, cited to Savova et al. (2019), to auto-suggest new links from the literature. Educational Overlay would develop a guided “walkthrough” mode in the viewer using scaffold anchors and narrative annotations. Enhanced Interoperability would support SBGN export and tighter alignment with Physiome Model Repository standards.

Taken together, these points define ApiNATOMY as both a representational formalism and an operational toolchain. Its template-based approach and scaffold anchoring enable rapid construction of multi-scale, topologically accurate physiological circuit maps, while the JSON schema and JSON-LD context ensure FAIR compliance and broad interoperability with existing biomedical ontologies (Kokash et al., 19 Sep 2025).