A machine learning model presented at AACR 2026 showed strong performance in predicting the tissue of origin for cancers of unknown primary (CUP) using DNA methylation data. Researchers from Kindai University reported ~95% correct cancer-type identification in a test cohort and 87% accuracy in an independent validation cohort. The study emphasized marker efficiency, selecting about 1,000 CpG regions from hundreds of thousands across the genome while maintaining performance. The investigators developed the model using methylation data from nearly 7,500 patients across 21 cancer types, drawn from The Cancer Genome Atlas and other public datasets. CUP often carries poor outcomes because patients receive broad chemotherapy instead of site-specific targeted therapy. The research aimed to help resolve origin even when primary tumors are not visible. As a limitation, the model was trained on cancers with known origins rather than true CUP cases, which will shape what validation work regulators and clinicians will demand next.