DNA methylation classifier predicts azacitidine response in MDS: AUC 0.88 in external cohort, Sylvester

Researchers at Sylvester Comprehensive Cancer Center, part of the University of Miami Miller School of Medicine, and partner centers analyzed 98 patients with myelodysplastic syndromes (MDS) and reported that DNA methylation at diagnosis helps predict response to azacitidine (AZA). The multicenter study was published in the Journal of Clinical Investigation and announced on November 9, 2026 jci.orgJournal of Clinical Investigationjci.orgOpen source ↗. It is titled "DNA methylation status at diagnosis in MDS predicts response to azacitidine".

MDS is a group of disorders in which bone marrow does not produce healthy blood cells normally. Altered DNA methylation is a hallmark of the disease, and DNA methyltransferase inhibitors are used for treatment. Reliable biomarkers of response have been lacking, so doctors cannot dependably predict which patients will respond.

Epigenetic clusters and methylation patterns

The team combined genomic, epigenomic and transcriptomic data from the 98 patients and grouped them by hierarchical clustering. MDS was characterized by widespread DNA hypomethylation affecting distal regulatory elements. The analysis identified seven epigenetic clusters that correlated with distinct molecular drivers.

Cluster VI had a low mutational burden and an AZA response rate of 71% (P ≤ 0.01) jci.orgJournal of Clinical Investigationjci.orgOpen source ↗. According to the authors, some of the strongest predictive signals came from regulatory regions outside the parts of DNA studied traditionally.

Classifier performance

The authors developed random forest classifiers to predict AZA response. They validated them with internal stratified testing and with an independent external cohort. Transcriptional profiles alone failed to distinguish responders. A classifier based on DNA methylation reached an area under the curve (AUC) of 0.82.

An integrative model combining DNA methylation, gene expression, mutations and clinical parameters reached an AUC of 0.93 in internal validation and 0.88 in the external cohort. The authors concluded that epigenetic signatures at distal genomic elements gave better predictive power for AZA response than promoter-centered or transcriptional analyses.

Maria "Ken" Figueroa, M.D., associate director of translational research at Sylvester and co-lead of the study, said the findings could provide information that conventional approaches have not captured. Mikkael Sekeres, M.D., M.S., chief of hematology at Sylvester and the other co-lead, said further research is needed before the findings can guide routine care. The first author is Qin Yang, Ph.D., of the Miller School.