140 Years of mathematical modeling in oncology through AI-assisted curation

Constructing a comprehensive overview of any scientific field requires accurate literature selection, yet conventional keyword-based searches are susceptible to false positives. This problem is magnified in growing or interdisciplinary fields such as mathematical modeling in oncology that contain a rich but heterogeneous body of literature. Here, a generalizable, context-enriched artificial intelligence pipeline based on large language models (LLMs) curates large scientific corpora according to a user-defined field: mathematical modeling in oncology (>35k publications). Benchmarking against expert evaluation demonstrates high accuracy (ROC AUC[~]0.95) and agreement with human judgement (correlation[~]0.68), outperforming zero-shot LLM curation Analysis of the curated corpus ([~]14k) suggests that Mathematical Oncology s distinct from either Systems Biology and Pharmacokinetics/Pharmacodynamics despite employing overlapping methods. Co-occurring citation network analysis defines nine research clusters focused on a range of applications including drug delivery, optimal control, stochastic modeling, tumor microenvironment, radiation, cancer evolution, and spatial multiscale modeling. Significance StatementA generalizable, context-enriched artificial intelligence pipeline accurately curates large scientific corpora Analysis of the curated dataset applied to the se of mathematics n oncology provides comprehensive view of mathematical modeling in oncology across 140 years, revealing its shift from fundamental cancer biology towards therapeutic modelling.