Benchmarking of methods to analyse data derived from GBS-MeDIP

Genotyping-by-Sequencing with Methylated DNA Immunoprecipitation (GBS-MeDIP) is an emerging method for cost-effective DNA methylation analysis. However, due to its unique sequencing output, conventional bioinformatics pipelines used for RNA-seq and MeDIP-seq are not fully adequate for analyzing GBS-MeDIP data. Selecting the appropriate statistical methods for differential methylation analysis remains a challenge, as existing approaches may introduce bias or false positives. We benchmarked multiple statistical methods for analyzing GBS-MeDIP data using previously generated datasets from chickens, dogs, and pigs. FeatureCounts was identified as the most reliable tool for count matrix generation, outperforming MEDIPS, which introduced biases in count estimation. For differential methylation analysis, we evaluated EdgeR, limma, DESeq2, and the Mann-Whitney test. Our results demonstrated that Mann-Whitney provided the lowest false positive rate and highest true positive rate, outperforming both EdgeR, DESeq2, and limma. EdgeR’s quasi-likelihood method exhibited a high false positive rate, making it unsuitable for GBS-MeDIP analysis. Our findings highlight that GBS-MeDIP data should not be analyzed using standard RNA-seq or MeDIP-seq pipelines, as these approaches lead to statistical artifacts. Instead, we recommend featureCounts for count matrix creation and Mann-Whitney for differential methylation analysis, ensuring accurate detection of differentially methylated windows. This study provides a bioinformatics framework for analyzing GBS-MeDIP data, minimizing biases and improving reliability in epigenomic research.