Structured Schemas for Provenance-Rich, LLM-Assisted QSP Model Calibration.

Area of Science:

• Pharmacology
• Computational Biology
• Data Science

Background:

• Quantitative systems pharmacology (QSP) models rely on literature data for calibration.
• Manual data curation is inconsistent and prone to errors.
• Large language models (LLMs) can hallucinate values and fabricate citations when extracting data.

Purpose of the Study:

• To introduce MAPLE (Model-Aware Parameterization from Literature Evidence), a novel framework for QSP model parameterization.
• To enhance the accuracy, consistency, and provenance of data extraction from literature for QSP models.
• To create a collaborative interface between LLMs and modelers for robust data curation.

Main Methods:

• MAPLE utilizes structured validation schemas (SubmodelTarget and CalibrationTarget) to separate data extraction from modeling decisions.
• Schemas ensure full provenance recording for every extracted value.
• Targeted validators automatically detect and correct characteristic LLM errors, resolve DOIs, and execute code.

Main Results:

• MAPLE successfully extracted and curated 37 SubmodelTargets and 45 CalibrationTargets for a pancreatic ductal adenocarcinoma QSP model.
• Automated validators triggered 50 retries before human review, ensuring data accuracy.
• 11 of 19 parameters were supported by multiple independent sources, and all values had verified citations.

Conclusions:

• MAPLE significantly improves the reliability of literature data extraction for QSP models by integrating LLMs with structured validation.
• The framework enhances reproducibility by recording modeler reasoning and enabling independent checks.
• MAPLE facilitates a synergistic collaboration between AI and human expertise in computational modeling.