metilene³: identifying DMRs across multiple conditions with auto-classification

DNA methylation is a critical epigenetic mark across numerous species, and identifying differentially methylated regions (DMRs) is essential for understanding genome regulation. Most existing DMR detection methods require predefined sample conditions, limiting the discovery of new epigenetic patterns, especially when group identities are unknown or uncertain, as is common in clinical settings. Additionally, only a very few approaches enable comparisons across multiple conditions. To address this significant gap, we present metilene³, a method for rapid, multi-condition DMR detection that operates in both supervised and unsupervised modes, using user-provided labels or autonomously clustering unlabeled samples. By segmenting the genome based on multiple pairwise methylation difference signals, metilene³ enables sample classification and DMR-anchored inference of epigenetic relationships. Using simulated and diverse human datasets, we show that metilene³ accurately detects DMRs, robustly clusters samples, and holds the potential to reveal new regulatory elements and sample stratifications. Specifically, in a pancreatic tissue dataset, metilene³ identifies DMRs enriched for key transcription factors involved in pancreatic cancer development, hinting towards an altered NFKB-NFAT regulatory program. Together, metilene³ provides a fast, interpretable framework for exploring heterogeneous methylomes and discovering epigenetic patterns across complex biological and clinical datasets.