1Optimal treatment of brain metastases is often hindered by limitations in diagnostic 2 capabilities. To meet these challenges, we generated genome-scale DNA methylomes of the 3 three most frequent types of brain metastases: melanoma, breast, and lung cancers (n=96). 4Using supervised machine learning and integration of multiple DNA methylomes from 5 normal, primary, and metastatic tumor specimens (n=1,860), we unraveled epigenetic 6 signatures specific to each type of metastatic brain tumor and constructed a three-step DNA 7 methylation-based classifier (BrainMETH) that categorizes brain metastases according to the 8 Brain metastases (BM) are the most common intracranial neoplasm in adults and are 1 the next frontier for the management of metastatic cancer patients. Large population-based 2 studies have shown that 8-10% of cancer patients develop brain metastases, with this 3 proportion increasing up to 26% when autopsy studies were included 1-5 . Lung cancer, breast 4 cancer, and cutaneous melanoma account for the vast majority (75-90%) of secondary 5 neoplasms in the brain 1-4 . Treatment options for BM include surgery, whole-brain 6 radiotherapy, stereotactic radiosurgery, and systemic-drug therapy, such as immunotherapy 6 . 7While systemic chemotherapy has limited efficacy, targeted therapies have recently shown 8 promise for the management of patients with specific subtypes of cancer 6 . These tailored 9therapies have significantly affected treatment decision making for patients with breast 10 cancer BM (BCBM). For example, patients with human epidermal growth factor receptor 2 11 (HER2)-positive BCBM can be treated with anti-HER2 agents 7 and patients with estrogen 12 receptor (ER)-positive BCBM can be treated with endocrine agents, cyclin dependent 13 kinases 4 and 6 (CDK4/6) inhibitors, and the mechanistic target of rapamycin kinase (mTOR) 14 inhibitors 8 . As such, accurate diagnosis is essential to effectively treat patients with 15 metastatic brain tumors. 16 17 Diagnosis of BM is currently based on neuro-imaging and confirmed by anatomic 18 pathology examination. When appropriate, the diagnostic algorithm begins by distinguishing 19 BM from primary brain tumors using histologic features guided by the clinical and radiologic 20 information 9 . Then, to identify the tissue of origin, morphological evaluation is supplemented 21 by several immunohistochemistry (IHC) markers including thyroid transcription factor (TTF-22 1), chromogranin and synaptophysin for lung cancer BM (LCBM); GATA3 binding protein 23 (GATA3), mammaglobin, gross cystic disease fluid protein 15 (GCDFP-15) and ER for 24 BCBM; and human melanoma black 45 (HMB45), melanoma antigen recognized by T-cells 1 25 4 (Melan A/MART-1), SRY-Box 10 (SOX-10), and the S100 calcium binding proteins (S-100) 1 for melanoma BM (MBM) 9,10 . However, a major limitation in achieving an accurate 2 pathological diagnosis is the often poor differentiation and/or limited availability of metastatic 3 brain tumor tissues to evaluate the complete panel of IHC mark...