Method Profile
Two structured record types link literature measurements to parameter priors or complete-model endpoints, with automated checks followed by scientific curation.
Make literature-derived calibration constraints and the modeler’s translation assumptions inspectable and reproducible.
One pancreatic ductal adenocarcinoma QSP model, with isolated experimental targets and baseline, GVAX ± nivolumab, and progression endpoints.
MAPLE v0.1.0 drafts SubmodelTarget and CalibrationTarget records, validates references/values/units/code, records human revisions and generates joint inference code for shared parameters.
Corpus
37 SubmodelTargets and 45 CalibrationTargets: 26 baseline, 18 treatment-related and 1 progression endpoint.
Provenance fields
Reported values, snippets, DOI/title, units, source context, forward-model or observable mapping, priors and uncertainty rationale.
Separate correct extraction from scientific applicability. Source-text snippet verification was selective during curation, not part of every automated gate.
Compare initial and curated structured records and instrumented retry traces in a single disease/model application.
37 parameter records; 45 full-model targets; detailed retry tracing for 18 batch SubmodelTarget runs; 18 surviving batch CalibrationTargets have paired drafts, with 27 additional interactive records.
Raw batch records versus curated records; automatic gate output versus modeler decisions.
18 monitored extractions triggered 50 retries, with no first-attempt pass. Human curation changed forward-model type in 65%, priors in 46% and source relevance in all 37 parameter records. Nineteen unique parameters were inferred; 11 have multiple supporting targets.
MAPLE structures literature-derived data for quantitative systems pharmacology calibration. A pancreatic-cancer case shows why automatic checks and human judgment have separate roles: 18 monitored extractions required 50 retries, and scientific fields still needed substantial revision.
