Agentic Framework for Epidemiological Modeling

TL;DR

EpiAgent addresses the need for scalable, interpretable, and scenario-consistent epidemic simulators by treating simulator construction as an agentic, program-synthesis problem. It introduces a flow-graph intermediate representation, retrieval-augmented graph synthesis, and a multi-agent verification and validation loop that enforces structural and epidemiological constraints before code generation. Empirical results show accurate calibration, robust counterfactual reasoning, and accelerated convergence to valid, interpretable simulators across COVID-19 scenarios and behavioral baselines. The framework promises scalable, expert-scripted epidemiological modeling suitable for evolving public health decisions while maintaining interpretability and mechanistic integrity.

Abstract

Epidemic modeling is essential for public health planning, yet traditional approaches rely on fixed model classes that require manual redesign as pathogens, policies, and scenario assumptions evolve. We introduce EPIAGENT, an agentic framework that automatically synthesizes, calibrates, verifies, and refines epidemiological simulators by modeling disease progression as an iterative program synthesis problem. A central design choice is an explicit epidemiological flow graph intermediate representation that links scenario specifications to model structure and enables strong, modular correctness checks before code is generated. Verified flow graphs are then compiled into mechanistic models supporting interpretable parameter learning under physical and epidemiological constraints. Evaluation on epidemiological scenario case studies demonstrates that EPIAGENT captures complex growth dynamics and produces epidemiologically consistent counterfactual projections across varying vaccination and immune escape assumptions. Our results show that the agentic feedback loop prevents degeneration and significantly accelerates convergence toward valid models by mimicking professional expert workflows.

Figures (17)

• Figure 1: Data flow in EpiAgent has the following structure: Prompt$\;\leftrightarrow\;$Flow Graph$\;\rightarrow\;$Simulator Code$\;\rightarrow\;$Results. The flow graph is an abstract representation of the epidemic model being learned, and supports easier verification. The flow graph is then transformed into the actual epidemic model and is calibrated, and fully verified. The components here map to those in the architecture of EpiAgent, shown in Figure \ref{['fig:pipeline']}.
• Figure 2: Agentic pipeline for epidemic scenario modeling and ensemble simulation. Natural-language scenarios are augmented with domain knowledge (§\ref{['ssec:method-scenario']}), to produce prompts that guide flow-graph synthesis (§\ref{['ssec:method-graph']}). Generated graphs are iteratively verified to enforce valid compartmental structure and transitions. Given a verified graph and scenario description, an LLM planner instantiates executable simulator code with automated error recovery (§\ref{['ssec:method-simulator']}). Simulators are calibrated on observed data and evaluated as a scenario ensemble (§\ref{['ssec:method_train']}). A multi-agent verification and validation stage enforces epidemiological and scenario-consistency constraints, retaining only structurally and behaviorally valid models (§\ref{['ssec:method_vnv']}).
• Figure 3: Cumulative COVID-19 infections aggregated at the national level across six scenarios from midas-covid19-scenario-modeling-hub. State-level trajectories are normalized by population and aggregated to produce a single nationwide infection and death trajectory.
• Figure 4: Error propagation from incorrect graphs to epidemic projections: may yield plausible outcomes (b) or incorrect trajectories (a)–demonstrating the necessity of structural verification.
• Figure 5: Iterative refinement of epi models under agentic verification and feedback. Subsequent feedback corrects structural violations and progressively refines model expressiveness.