LLM-based Biomedical Article Figure Annotation for Ontology Development

Abstract

There has been remarkable research and development across various fields using Large Language Models (LLMs) in recent years. An integrated understanding across granularities, from molecules to organisms, is essential in biomedicine. We have developed biomedical ontologies for the development based on manual annotation. However, manual annotation is costly and requires more effort to update. In this study, we report on using Large Language Models (LLMs) for annotation from figures in articles, exemplified by typical cellular senescence. Next, we describe ontology mapping for the systematization of knowledge, and finally, we discuss the evaluation of this approach by comparing it to manual annotations.