Boolean models — sources and tooling landscape

SCOPING PIPELINE

Sources and tooling landscape

Companion to BOOLEAN_MODELS. Everything below was checked
from this container on 2026-09-26 (API calls, downloads, package installs); facts
that could not be verified are marked as such.

What a Boolean model is, for this project

A Boolean (logical) model is a set of variables, each with an update rule that is a
Boolean function of its regulators; the regulatory graph is the signed dependency
graph of those rules. Its behaviour is the set of attractors (fixed points and
cyclic/complex attractors) under an update scheme (synchronous, or asynchronous as
used by GINsim and most published models). The SBML Level 3 qual package is the
interchange standard (PMID:24321545), and
BoolNet .bnet text (target, factors) is the de facto lightweight format.

The relevance to a curation KB: a module's connections graph is a regulatory graph
with signs, so a Boolean network is the cheapest dynamical reading of a module, and
the published Boolean models are hand-curated signed wiring for many of the same
pathways.

External sources

Source What it holds Access from here Boolean-model relevance
SIGNOR 3.0 (PMID:36243968) Signed, mechanism-typed causal relations between proteins/complexes/families/phenotypes, curated with PMID + sentence; about 130 named pathways getPathwayData.php?pathway=SIGNOR-EGF&relations=only returns a TSV; ?list lists pathways. Works from this container. CC BY 4.0. Not a Boolean model, but exactly the signed-edge layer a Boolean regulatory graph is built from. SIGNOR's own Boolean pipeline goes SIGNOR → CaSQ → SBML-qual (PMID:32403123). Snapshot: models/boolean/signor/SIGNOR-EGF.tsv (102 rows, 61 unique signed pairs; P38 143 rows, SAPK-JNK 72, NFKBC 44, TA 68).
BioModels (PMID:31701150) Published models incl. SBML-qual logical models REST search works via https://www.ebi.ac.uk/biomodels/search?query=...&format=json with an Accept: application/json header (the bare URL redirects to www.biomodels.org and loses the format). A free-text Boolean query returns 37 models; the modellingapproach facet counts 16 logical model and 13 boolean model entries (overlapping). Facet-filtered queries with quoted values were rejected by the proxy path and need the web UI or a different encoding. Model files download from /biomodels/model/download/<id>?filename=.... Manually curated SBML-qual examples: BIOMD0000000562 (Chaouiya 2013 EGF/TNFα demo, 28 qualitative species), MODEL2106070001 (Montagud 2022 prostate cancer), MODEL1611180000 (Verlingue 2016 S-phase entry), MODEL2006170001/MODEL2006170002 (Regan/Sizek cell-cycle–apoptosis models), MODEL2304070002 (Ruscone 2023 tumour invasion).
BBM — Biodivine Boolean Models (Pastva et al., bioRxiv 10.1101/2023.06.12.544361) 285 logically consistent real-world Boolean networks harvested from GINsim, Cell Collective, BioModels and papers, each as .bnet/.aeon/.sbml with metadata and a summary.csv raw.githubusercontent.com/sybila/biodivine-boolean-models/main/models/… works from here (github.com HTML pages do not). The most convenient corpus: uniform format, inputs explicit, provenance JSON. MAPK-relevant entries: 070 MAPK-CANCER-CELL-FATE (Grieco 2013), 089–091 its reductions, 018 EGFR-ERBB, 136 EGF-TNFα, 012/032/080 T-cell receptor, 020/111/138 apoptosis, 219 WNT-PI3K-AKT. Snapshot: models/boolean/bbm-070-mapk-cancer-cell-fate/.
GINsim model repository (PMID:29971008) The canonical multi-valued/Boolean logical models (.zginml) ginsim.org TLS handshake fails from this container; use BBM's mirror or bioLQM offline. Original source of many BBM entries (e.g. Grieco 2013 = ginsim.org/node/173).
Cell Collective (PMID:22871178, REST API PMID:26589448) Community-built logical models with per-edge literature annotation api.cellcollective.org is not reachable through the proxy (502). Its edge-level citations would be the natural counterpart to module evidence; reachable only via BBM copies for now.
GO-CAM (in-repo gocams/, PMID:31548717) 2,036 cached production causal-activity models with causal_associations (RO relations such as directly positively regulates, provides input for) Local cache; gocams/index.tsv joins gene products to activities. A GO-CAM is already a signed causal graph over activities and translates by the same rule as a module (RO relation → sign). Human MAPK-relevant models exist (e.g. 65d7e4ac00001732, 685de18700001720) but are receptor-specific slices rather than the whole cascade.
Reactome (in-repo reactome/) Reaction-level pathway diagrams; SBML-qual export via CaSQ-style tools exists in the literature Local cache. Reaction graphs are not regulatory graphs; a Boolean reading needs the CaSQ transformation (AND reactants, OR activators, AND-NOT inhibitors). Out of scope for the first pass.

Tooling

All Python packages below installed into a scratch virtualenv from PyPI through the
container proxy and imported successfully; uv run --with biodivine-aeon also works
as an ephemeral overlay on the project environment (this is how the demo runs).

Tool Role Notes
biodivine-aeon (AEON.py, PMID:36102786) Symbolic (BDD) attractor analysis of asynchronous Boolean networks; reads/writes .bnet, .aeon, SBML-qual; infers a signed regulatory graph from rules (infer_valid_graph) Used by the demo. Free inputs (rules omitted) are treated as parameters, so fix them to constants before computing attractors or every input combination is folded into one "coloured" attractor. Attractor projection is done symbolically by intersecting with mk_subspace. Computes the 53-variable Grieco model in ~1 s.
mpbn Most-permissive Boolean network semantics (Paulevé lab); attractor/reachability via ASP Installs (pip install mpbn), needs clingo wheel; not used yet. Reference to cite when the most-permissive semantics is adopted.
PyBoolNet (PMID:27797783) Trap spaces, model checking (NuSMV), primes Installable; heavier native deps.
bioLQM (PMID:30510517) / GINsim Java toolkit: format conversion (zginml ↔ SBML-qual ↔ bnet ↔ BoolNet), Booleanisation of multi-valued models Reference converters; would need a JVM here.
CoLoMoTo notebook (PMID:29971009) Docker image bundling all of the above with reproducible notebooks The right vehicle for a shareable, reproducible analysis once the pipeline stabilises.
MaBoSS (PMID:28881959) Stochastic continuous-time Boolean simulation (probabilities of phenotypes) For phenotype-probability read-outs (e.g. proliferation vs apoptosis) rather than attractor enumeration.
CaSQ (PMID:32403123) Map (SBGN/CellDesigner, SIGNOR) → Boolean model with the default rule AND reactants, OR activators, AND NOT inhibitors The default-rule convention adopted by module_boolean.py; the tool itself targets SBGN maps.
boolean.py Boolean expression algebra Not needed; module_boolean.py carries a 60-line parser/evaluator sufficient for .bnet rules.

Formats