Ontology-Guided, Verifiable In-Silico Experiments on Lumped-Parameter Cardiovascular Models: Semantic Connection Checking, SHACL Validation and a Receipt-Verified Language-Model Agent

Composing physiological models from independent components fails at the joints: same-unit variables can carry different meanings, metadata can be correct while the attached numbers are wrong, and a language model that writes the report can state results never computed. We present an open framework treating each separately. The core is a deterministic six-compartment closed-loop cardiovascular model integrated as a hybrid system of valve regimes with runtime conservation, positivity, valve and settling audits. A semantic layer annotates 34 variables with label-checked Ontology of Physics for Biology properties and Uberon/FMA anatomy, exports named-graph RDF and provides a fail-closed connection validator requiring agreement of property, anatomy, bearer, reference, temporal context and flow orientation before a value is transferred. In a 12-fixture fault-injection benchmark it rejects eight of nine unsafe connections an unguarded baseline accepts; its retained blind spot, correct left-ventricular metadata on right-ventricular data, passes with errors up to 129.8 mmHg. An evidence-first review of the Smith et al. (2004) CellML model produced a variable dictionary, anatomical node specification, SHACL shapes and PROV-O records; on the model’s published SemSim annotation they report 22 violations, 7 warnings and 46 informational results. A tool-calling language-model agent plans baseline-versus-intervention experiments under a three-layer scope guard; it chooses only runs and metric paths, and every sentence and number in its report is written by code from tool receipts and re-verified. In a 27-case evaluation three providers passed every case; all 14 out-of-scope clinical requests were refused before any model call. Nothing is clinically validated; parameters are exploratory, claim boundaries stated. Code, evidence (redacted where required), shapes, provenance and traces are public.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-13
DOI
https://doi.org/10.5281/zenodo.24182405
Primary Topic
Scientific Computing and Data Management
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Ontology-Guided, Verifiable In-Silico Experiments on Lumped-Parameter Cardiovascular Models: Semantic Connection Checking, SHACL Validation and a Receipt-Verified Language-Model Agent

Sungyeob Lim
Zenodo (CERN European Organization for Nuclear Research)
Scientific Computing and Data Management
preprint

Ontology-Guided, Verifiable In-Silico Experiments on Lumped-Parameter Cardiovascular Models: Semantic Connection Checking, SHACL Validation and a Receipt-Verified Language-Model Agent

Sungyeob Lim
preprint en

Abstract

Composing physiological models from independent components fails at the joints: same-unit variables can carry different meanings, metadata can be correct while the attached numbers are wrong, and a language model that writes the report can state results never computed. We present an open framework treating each separately. The core is a deterministic six-compartment closed-loop cardiovascular model integrated as a hybrid system of valve regimes with runtime conservation, positivity, valve and settling audits. A semantic layer annotates 34 variables with label-checked Ontology of Physics for Biology properties and Uberon/FMA anatomy, exports named-graph RDF and provides a fail-closed connection validator requiring agreement of property, anatomy, bearer, reference, temporal context and flow orientation before a value is transferred. In a 12-fixture fault-injection benchmark it rejects eight of nine unsafe connections an unguarded baseline accepts; its retained blind spot, correct left-ventricular metadata on right-ventricular data, passes with errors up to 129.8 mmHg. An evidence-first review of the Smith et al. (2004) CellML model produced a variable dictionary, anatomical node specification, SHACL shapes and PROV-O records; on the model’s published SemSim annotation they report 22 violations, 7 warnings and 46 informational results. A tool-calling language-model agent plans baseline-versus-intervention experiments under a three-layer scope guard; it chooses only runs and metric paths, and every sentence and number in its report is written by code from tool receipts and re-verified. In a 27-case evaluation three providers passed every case; all 14 out-of-scope clinical requests were refused before any model call. Nothing is clinically validated; parameters are exploratory, claim boundaries stated. Code, evidence (redacted where required), shapes, provenance and traces are public.

Zenodo (CERN European Organization for Nuclear Research)
Soonchunhyang University (KR)
Scientific Computing and Data Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon