Researchers presented methylation-based machine learning models that aim to identify the tissue of origin for cancers of unknown primary (CUP) by using CpG DNA methylation signatures. The work targets a clinical gap where many CUP patients receive broad chemotherapy because the primary site can’t be identified with standard pathology and IHC. At AACR, Guardant Health described a methylation machine learning classifier intended to guide precision treatment selection, while investigators from Kindai University reported a CpG-based model that showed ~95% correct identification across the test cohort and 87% accuracy in an independent validation set spanning multiple cancer types. These efforts focus on shrinking the fraction of patients who remain without actionable stratification—while also reducing the reliance on larger panels or longer diagnostic workflows. Further validation will hinge on prospective evidence that tissue-of-origin calls improve outcomes.