Predictive Epitranscriptomics: Computational Identification of m6A Methylation Patterns Associated with Future β-Cell Dysfunction and Hyperglycemic Transition

Objective To develop a computational framework integrating m6A methylation profiles with machine learning to identify patterns predictive of future β-cell dysfunction and hyperglycemic transition. Methods We performed a multi-phase bioinformatics analysis of transcriptome-wide m6A and RNA-seq data from human pancreatic islets across normoglycemic, prediabetic, and T2DM states. Differential methylation and expression analyses were conducted using established pipelines. Machine learning models were trained and validated on m6A features, transcript expression, and clinical variables. Results m6A methylation patterns robustly distinguished disease states, outperforming transcriptomic profiles alone. Hypomethylation of key β-cell genes (PDX1, MAFA, INS) and insulin signaling pathway components was strongly associated with β-cell dysfunction. Machine learning models achieved high accuracy (AUC-ROC 0.94) in predicting T2DM risk, with m6A features being the most influential predictors. Longitudinal analysis revealed progressive m6A hypomethylation preceding clinical hyperglycemia. Conclusion m6A methylation signatures serve as powerful biomarkers for early detection of β-cell dysfunction and hyperglycemic transition, offering a novel avenue for predictive medicine in DM.