PopGenAgent: An Agent-Mediated Workflow for Population Genetics Analyses

Abstract

Population genetics analyses typically require orchestrating multiple tools across heterogeneous tasks, from data curation to inference and visualization. In practice, a central bottleneck lies not only in reproducibility, but in maintaining consistency between intermediate diagnostics, analytical decisions, and downstream results as analyses evolve. Here we present PopGenAgent, an agentic workflow that formalizes population-genetic analyses as explicit, multi-step plans with declared inputs, outputs, and dependencies. Each step is instantiated from curated tool and visualization templates, while all intermediate artefacts are preserved within a unified, inspectable context. An interactive agent interface supports iterative plan construction and revision, and leverages accumulated outputs to assist interpretation and report drafting. In a case study of 26 populations from the 1000 Genomes Project, starting from a filtered PLINK dataset, PopGenAgent coordinated analyses including ROH, LD decay, PCA, ADMIXTURE, TreeMix, and f3. By structuring analyses as explicit, revisable plans with preserved intermediate states, PopGenAgent provides a transparent and extensible environment for conducting population-genetic analyses.