A multimodally extracted oxide nanomaterial in vitro cytotoxicity dataset

Park-OxideNM2026 is a structured oxide nanomaterial in vitro cytotoxicity dataset generated for the study "Automated Large-Scale Extraction of Nanotoxicity Data Using Multimodal Large Language Models". The dataset was constructed by applying a multimodal large language model workflow to oxide nanotoxicity literature, integrating information from full article text, figures, tables, captions, legends, and Methods sections to recover material, physicochemical, experimental, and toxicological assay information.The dataset contains 16,786 gap-filled records extracted from 465 articles. Each record represents an oxide nanomaterial cytotoxicity assay condition and includes harmonized descriptors for material properties, physicochemical characteristics, assay context, exposure conditions, and toxicity outcome.