ATLAS process

From literature to SBML decision engine

A staged pipeline for building, validating and updating mechanistic models with transparent human review.

1. Evidence extraction

Mine publications, labels, trial reports and model papers; identify mechanisms, endpoints, assumptions and quantitative anchors.

2. Model reconstruction

Rebuild the mechanistic structure as species, compartments, reactions, rules, events and observable equations.

3. SBML assembly

Encode in SBML Level 3 Version 2 where possible, with file naming, provenance and reproducible simulation assets.

4. Calibration

Fit selected parameters to curated targets and hold out data for predictive checks where feasible.

5. V&V and uncertainty

Run structural checks, numerical tests, RMSE/holdout criteria, sensitivity, identifiability and Bayesian summaries as appropriate.

6. Reports and release

Publish the model page, package artefacts, update the catalogue and define the free/subscription boundary.

Quarterly updates

Each active model can be refreshed on a quarterly cadence: new literature, new labels, new trials, revised assumptions, validation report deltas and partner-specific scenario packs.