A comprehensive evaluation of computational models for identifying methylation variability in DNA methylation arrays.

DNA methylation (DNAm) is a key epigenetic modification that regulates gene expression and is involved in biological processes such as development, disease progression, and responses to environmental factors. While differential methylation analysis, which focuses on average methylation differences, is widely used in epigenome-wide association studies, variable methylation (VM) analysis is gaining recognition for its ability to uncover dynamic molecular processes underlying various phenotypic outcomes. VM analysis, which focuses on variability in methylation patterns, is particularly valuable for detecting subtle epigenetic changes that may signal early disease stages or complex mechanisms, such as those observed in cancer, where these changes reflect shifts in gene regulation that may not be captured by average methylation levels. Computational tools for VM analysis utilize a variety of approaches, broadly categorized into feature selection-based, joint location-scale test, and variance-based methods, making tool selection challenging. In this study, we conducted a comprehensive evaluation of eight tools that can be used for VM analysis to provide empirical evidence and guidance for researchers in choosing most appropriate tool. Using extensive simulations, we assessed performance of these tools based on various simulation scenarios and evaluations metrics. Our findings indicate that most tools controlled false discovery rate well, except for DiffVar in certain scenarios. All tools had limited power under a small sample size and moderate effect size setup which changed as both sample and effect sizes increased. Across all simulation scenarios and evaluation metrics, JLSsc demonstrated favorable performance as compared to other models in identifying VM in DNAm microarray data.