Skip to content

Academic Research Workbench

IINTS-AF can add a reviewable academic metadata layer to a completed simulation run. The goal is practical reproducibility: another researcher should be able to identify the software, configuration, seed, artifacts, checksums, and references without relying on a screenshot or an AI summary.

Scope

An academic package is not peer review, ethical approval, privacy clearance, clinical validation, or medical-device certification. It does not upload data. Review every artifact before sharing it.

Create A Package

iints research academic-bundle results/my_run \
  --title "Baseline T1D reproducibility run" \
  --creator "Researcher Name" \
  --orcid "https://orcid.org/0000-0000-0000-0000" \
  --license "CC-BY-4.0"

To associate references explicitly, repeat --source-id:

iints research academic-bundle results/my_run \
  --source-id hovorka_2004_nmpc_t1d \
  --source-id attd_2019_time_in_range

When no source IDs are supplied, the exporter makes only conservative associations that can be inferred from run metadata and artifacts. These associations are candidates for human review, not proof that a paper validates the implementation.

The run-artifact licence defaults to NOASSERTION. The SDK code is Apache-2.0, but that software licence is not silently applied to real-data-derived CSV files, reports, or other research artifacts. Choose a data/output licence only when you have the right to do so.

The same operation is available in both desktop workbenches from the Results or Reproducibility package area.

Generated Files

File Purpose
ro-crate-metadata.json RO-Crate 1.2 JSON-LD with run, software, file, checksum, creator, licence, and source entities
academic_audit.json machine-readable checks for metadata, seed, configuration, manifest, revision, sources, attribution, licence, artifact inventory, and basic privacy markers
academic_sources.json exact evidence-registry snapshot, registry hash, selected sources, and selection method
ACADEMIC_BUNDLE.md short human review guide stored with the run

Existing experimental files are not copied or rewritten. The exporter reads them to calculate SHA-256 checksums and then writes the four metadata files beside them.

Readiness Status

Status Meaning
ready all implemented required, recommended, and review checks passed
needs_review required metadata exists, but one or more recommended or human-review checks remain
incomplete a required artifact or metadata field is missing

The score measures only the implemented checklist. A high score does not mean that a model is physiologically accurate or that a dataset may legally be shared.

Academic Integration Levels

The app labels every external resource by maturity so a portal link cannot be mistaken for a functioning scientific integration.

Level Meaning Current examples
Integrated the SDK calls a defined local/API workflow and writes a reviewable artifact RO-Crate export; AlphaFold; SBML/libRoadRunner; COPASI; CellML/OpenCOR validation; FMI/FMPy; BindingDB
Partial a useful query or render exists, but full versioned import and validation are not complete GTEx expression, ChEMBL context, STRING networks, ClinVar context
Planned a scientifically useful boundary is documented but not implemented SED-ML protocol export, automatic Physiome repository import, structured PubMed capture
Portal the app opens an official allowlisted resource; no evidence is ingested RCSB PDB, UniProt, Human Protein Atlas, ClinicalTrials.gov, Zenodo

Standards Direction

  • FAIR4RS guides software citation, metadata, access, interoperability, and reuse. IINTS-AF is FAIR-oriented; it is not externally FAIR-certified.
  • RO-Crate 1.2 is implemented for run-level metadata and artifact inventory.
  • SED-ML Level 1 Version 5 is the planned portable description for model/simulation/task/output protocols. IINTS-AF does not yet claim SED-ML compatibility.
  • SBML Level 3 Version 2 is supported for safe structural inspection of local reference files. Optional independent execution uses libRoadRunner. Current Python patient models are not claimed to be SBML models.
  • BioModels is useful for model provenance and comparison. Automatic repository import is not implemented; local model files can be inspected only after the researcher has reviewed their source and licence.
  • Zenodo remains a manual publication route after privacy, licensing, and completeness review; the SDK never uploads automatically.

Evidence Rules

  1. Record the exact database, identifier, query, access date, and local artifact hash.
  2. Prefer primary publications and official database records over summaries.
  3. Keep structural confidence, variant assertions, expression evidence, and physiological parameters separate.
  4. Never convert AlphaFold pLDDT or PAE directly into pathogenicity, insulin sensitivity, or a dosing parameter.
  5. Never let external evidence or local AI silently alter the deterministic simulation configuration.
  6. Preserve failed runs, exclusions, software versions, and changed assumptions.
  7. Review direct and indirect identifiers before publishing any real-patient-derived data.
flowchart LR
    A["Freeze protocol and hypotheses"] --> B["Run with recorded seeds"]
    B --> C["Validate raw artifacts"]
    C --> D["Create academic package"]
    D --> E["Resolve audit findings"]
    E --> F["Review privacy and licences"]
    F --> G["Archive code, environment, and approved outputs"]

Use deterministic metrics and raw time series as the numerical authority. Local AI can help critique or explain a run, but its response is a review note and is not included as ground truth.

For independent external equation-model checks, use Mechanistic Reference Models. Reference execution remains separate from IINTS calibration so a successful solver run cannot silently change the virtual patient.

For sensitivity tasks, CellML validation, physical-device FMUs, and measured affinity evidence, continue with Cross-scale Reference Labs. These outputs remain separate evidence layers and require explicit mappings before comparison.