Loading Curie Briefs...
Getting the latest healthcare news for you
Getting the latest healthcare news for you
A new machine learning tool called metilene3 can detect DNA methylation patterns without needing pre-labeled samples, uncovering hidden disease subgroups. Tested on blood cells, glioblastomas, and pancreatic cancer, it revealed previously unknown biological relationships and regulatory processes. Researchers say it could open doors to new biomarkers and therapeutic targets.
Scientists from Berlin, Potsdam, and Jena have developed metilene3, a machine learning tool that analyzes DNA methylation — a key part of the epigenome that controls which genes are switched on or off — without requiring samples to be pre-labeled as "healthy" or "diseased." Published in Nature Communications, the method works in both supervised and unsupervised modes, making it especially powerful for complex clinical datasets where biological groupings aren't known in advance.
When put to the test, metilene3 delivered strong results across multiple cancer types. In glioblastoma data, it identified distinct molecular tumor subgroups and flagged samples with unusual biological properties. In pancreatic cancer tissue, it traced disease progression from healthy tissue through precancerous lesions to full tumors — and pinpointed DNA regions where transcription factors NF-κB and NFAT co-occur, which may play a role in cancer development (pending experimental verification).
Key Takeaways:
Why it matters: Epigenetic changes drive cancer, aging, and many other diseases — but studying them has been limited by the need for labeled data. Metilene3 removes that barrier, potentially accelerating biomarker discovery and opening new avenues for precision medicine.
