Skip to content

Repository files navigation

PhysioMap

PhysioMap is an ontology-grounded causal knowledge graph of human physiology. It represents physiological traits across molecular, cellular, tissue, organ, and whole-body scales and gives each relation type an explicit structural causal model semantics.

Release v1.1.1 Code license: BSD 3-Clause Data license: CC BY 4.0 Python 3.11+ Live viewer

The current release is PhysioMap v1.1.1. The release fixes the knowledge base, projection registry, causal model, evaluation inputs, generated results, and provenance needed to reproduce the accompanying paper.

Representation

The authoritative resource has three versioned parts:

OWL knowledge base + projection registry -> typed structural causal model

PhysioMap traits denote quantitative random variables. Its five content relation types constrain structural functions in different ways:

relation type structural causal model interpretation
causal influence a source variable enters a target mechanism as a non-constant direct dependence
production a process contributes an output term to the mechanism of its product
constitution constituent variables determine a variable of the whole through a constitutive map
quantitative identity a named variable equals a specified function of other variables
modulation a modulator changes the derivative of one variable with respect to another

Derivative signs form a separate abstraction of these quantitative constraints. The implemented first-order solver uses this abstraction to compute the sign of a steady-state response after an intervention. It returns + or - only when the available signs determine the response and returns ? otherwise. A separate gain-sensitivity query uses modulation relations.

Evidence and provenance are not PhysioMap content types and are not solver inputs. They document why each axiom was admitted, where it came from, and how it was reviewed. The distinction between content and construction evidence is described in docs/MULTISCALE.md. The released content review is documented in docs/EXPERT_GOLD_REVIEW.md.

Current release

content count
physiological traits 1,699
causal influences 2,270
production relations 85
constitutive constraints 4
quantitative identities 9
modulations 19
projected relation instances 2,387
largest causal feedback component 213 traits

The fixed-seed expert content review examined 83 projected relations across all five relation types. The reviewer accepted 69, flagged 12 for further investigation, and rejected 2. These counts describe the reviewed sample, not a map-wide accuracy estimate.

In the rare metabolic disease evaluation, the first-order solver made 171 determinate predictions among 866 directional gene-phenotype pairs, all concordant with the literature-adjudicated HPO-derived reference. The evaluation measures selective directional prediction and ranking within a closed lesion pool. It is not an independent clinical validation. The complete adjudicated pair set, including the primary intervention and supporting HPO classes for every row, is benchmarks/results/e1b_forward_pairs.tsv. The gene-stratified conditional randomization analysis of whether abstention tracks shortest-path errors is archived at immutable revision 79ff2fb. That revision contains the exact scripts/e2_baseline.py implementation and its human-readable and machine-readable results.

Install

git clone https://github.com/bio-ontology-research-group/physiomap.git
cd physiomap
uv sync --extra dev --extra analysis

Python 3.11 or later is required.

Quick start

Predict the directional consequences of reduced hepcidin:

uv run python -m physiomap_core.knockout hepcidin -

List the modulation relations available to the gain-sensitivity query:

uv run python -m physiomap_core.modulation --list

Run the test suite:

uv run pytest

Serve the interactive viewer locally:

uv run python web/export_data.py
uv run python -m http.server 8099 --directory web

Then open http://localhost:8099.

Reproduce the release

The complete release gate rebuilds the ontology and causal model, checks the projection, runs the test and evaluation contracts, verifies generated artifacts, and verifies deterministic reconstruction:

uv run python scripts/owl_scm_release_gate.py

The pinned HPO release inputs used by the rare-disease evaluation are stored under benchmarks/data/hpo-2026-02-16/. The gate verifies their compressed and expanded checksums before use.

When the separate manuscript repository is linked at paper/, the same command also compiles the paper and supplementary material. Submission maintainers can make those two additional checks mandatory:

uv run python scripts/owl_scm_release_gate.py --require-paper

The main evaluation outputs are generated by:

result command
directional rare-disease prediction uv run python scripts/e1b_eval.py --leakage-sensitivity
shortest-path comparison uv run python scripts/e2_baseline.py
precision-coverage comparison uv run --extra analysis python scripts/e2c_risk_coverage.py
inverse lesion ranking uv run python scripts/e4_diagnose.py
inverse-ranking comparisons uv run --extra analysis python scripts/e4b_diagnosis_baselines.py
manuscript figure and result macros uv run --extra analysis python scripts/generate_psb_rare_disease_figure.py

Generated outputs live under benchmarks/results/. The current acceptance criteria and interpretation of each evaluation are in docs/VALIDATION.md. Formal proofs and additional methods are in supplement/, including the Lean 4 formalization.

Repository layout

physiomap_core/   data model, projection consumers, and inference
ontology/         OWL construction and ontology grounding
projection/       versioned OWL-to-SCM projection patterns
release/owl-scm/  canonical v1.1.1 release artifacts and checksums
benchmarks/       evaluation inputs and generated results
scripts/          release, evaluation, and curation commands
tests/            automated test suite
docs/             current semantics, validation, and review documentation
supplement/       supplementary methods, proofs, and Lean formalization
web/              interactive viewer and exported release data

Scope and limitations

PhysioMap v1.1.1 supplies typed structural constraints and derivative signs, not complete quantitative functions, parameters, exogenous distributions, effect sizes, penetrance, thresholds, or dynamics. Numerical and population inference requires those additional inputs.

The first-order solver assumes re-equilibration at a locally stable state. Exact signed-determinant expansion is limited to feedback components of at most 16 traits by default; larger components use a conservative approximation. This cutoff is computational, configurable, and has no semantic significance. Modulation supports a separate qualitative gain query but was not evaluated in the rare-disease experiments.

Citation

Please cite:

Hoehndorf R, Schofield PN, Gkoutos GV. PhysioMap: an ontology-grounded causal knowledge graph of human physiology. Manuscript submitted to the Pacific Symposium on Biocomputing, 2026.

Machine-readable citation metadata is provided in CITATION.cff.

License

About

An ontology-grounded causal knowledge graph of human physiology with structural causal model semantics.

Topics

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages