MetaHarmonizer: robust biomedical metadata harmonization and a contamination control for inflated LLM performance on public benchmarks
Changchang Li, Abhilash Dahl, Kai Gravel-Pucillo, Kaelyn Long, Michele Waters, Ino de Bruijn, Sean R. Davis, Sehyun Oh
MetaHarmonizer improves metadata harmonization and addresses LLM contamination in biomedical data, outperforming existing methods on benchmark tasks.
bioRxiv · Jun 17, 2026 · PreprintBioinformatics