Talk2QSP: Deriving Executable Scenarios from Unstructured Literature via Human-in-the-Loop Agents

Abstract

Talk2QSP is an agentic, human-in-the-loop framework that translates literature-derived scenarios into simulation-ready Quantitative Systems Pharmacology (QSP) interventions. The pipeline comprises an Entity Definer, a Scenario Extractor, and a Scenario Mapper, supported by evaluation pipelines, tool-calling workflows, cost tracking, and observability and traceability across agent steps and state transitions. The system reduces the manual effort required to turn unstructured scientific literature into executable model interventions while keeping domain experts in the loop.

Publication
Preprint on bioRxiv — manuscript under review at ICLR 2026 GenBio Workshop
comments powered by Disqus

Related