Machine learning approaches reveal methylation signatures associated with pediatric acute myeloid leukemia recurrence.

Acute myeloid leukemia (AML) is a severe hematological malignancy characterized by high recurrence rates, especially in pediatric patients, highlighting the need for reliable prognostic markers. This study proposes methylation signatures associated with AML recurrence using computational methods. DNA methylation data from 696 newly diagnosed and 194 relapsed pediatric AML patients were analyzed. Feature selection algorithms, including Boruta, least absolute shrinkage and selection operator, light gradient boosting machine, and Monte Carlo feature selection, were employed to screen and rank methylation sites strongly correlated with AML recurrence. Incremental Feature Selection was performed to evaluate these results, and optimal subsets were identified using Decision Tree and Random Forest methods. Several important methylation features, such as modifications in SLC45A4, S100PBP, TSPAN9, PTPRG, ERBB4, and PRKCZ, emerged from the intersection of all feature selection algorithms. Functional enrichment analysis indicated these genes participate in biological processes, including calcium-mediated signaling and regulation of binding. These findings are consistent with existing literature, suggesting that identified methylation features likely contribute to AML progression through alterations in gene expression levels. Therefore, this study provides a valuable reference for enhancing recurrence risk prediction models in AML and clarifying disease pathogenesis, as well as offering broader insights into mechanisms underlying other major diseases.