DNA methylation is a key regulatory mechanism reflecting both short- and long-term biological stimuli. While it has been widely used to study aging through disease-associated methylation shifts, its potential for revealing tissue-specific shifts remains underexplored due to the lack of comprehensive reference atlases with correspondingly systematic analysis framework. To address this, we assemble the largest and most diverse atlas of healthy human tissue and cells profiled by 450K arrays, totaling 16,959 samples across 86 tissues and cell types. Using this resource, we introduce an ontology-aware classification framework that identifies robust CpG features linked to tissue and cell identity and incorporates known anatomical and functional relationships. Through minipatch learning, we distill 190 CpGs that support accurate multilabel classification and validate the approach with ontology-based label transfer to 31 unseen tissue and cell types.
Kim et al. (Sun,) studied this question.