Researchers from Kindai University presented an AI model at AACR 2026 that predicts the origin of cancers of unknown primary (CUP) by analyzing CpG DNA methylation patterns. The team reported that the model identified the cancer type in roughly 95% of cases in a test cohort and achieved 87% accuracy on an independent validation cohort. The approach selected about 1,000 CpG regions from across the genome to maintain performance while simplifying input requirements. The model was trained using methylation data spanning nearly 7,500 patients with 21 cancer types from The Cancer Genome Atlas and other public datasets. The researchers emphasized that the model was developed using cancers with known origins rather than true CUP cases, leaving open the need for further evaluation in real-world diagnostic workflows. Still, the work targets a key clinical gap where CUP patients often receive broad chemotherapy rather than site-directed therapy.