Limited knowledge exchange across biological subfields in agent-based modelling: an LLM-assisted bibliometric analysis
Agent-based models (ABMs) are widely used across the biological sciences. While applications differ between biological subfields, the mathematical and statistical methods underlying ABMs largely do not, raising the question of how much knowledge is exchanged between subfields. We address this question by developing a bibliometric pipeline that combines a database keyword search with coarse- and fine-grained classification by a large language model (LLM). We apply the pipeline to the Web of Science Core Collection, classifying 22,936 ABM articles into ecology, epidemiology, oncology, other-biology, or not-biology. Using citation patterns and author overlap as proxies for knowledge exchange, we find that 83%, 81%, and 69% of citations from ecology, epidemiology, and oncology articles, respectively, go to articles in the same subfield. Furthermore, 91% of authors with at least two articles across these subfields publish exclusively within one subfield. Both proportions far exceed those expected under random assignment of subfields (permutation tests, $p \leq 0.001$). Terminology also differs, with "individual-based model" used mainly in ecology and "agent-based model" in epidemiology, oncology, and outside biology. Taken together, our results suggest limited knowledge exchange between biological subfields in ABM research, leading us to discuss opportunities for increased exchange through conferences, journals, and training.