Benchmarking and Quality Control of Quantitative Accuracy in Next-Generation Sequencing-based DNA Methylation Profiling

Abstract Biomedical applications of DNA methylation rely on accurate next-generation sequencing (NGS)-based DNA methylation profiling. However, routine quality metrics and limited reference coverage leave it unclear whether NGS assays provide the reliable measurements required for robust downstream inferences. Here we benchmarked 12 real-world NGS assays across four methods, alongside five microarray and nanopore assays as external benchmarks. First, we assessed quantitative performance across three dimensions, including absolute methylation accuracy, biological order preservation, and the magnitude of relative methylation differences. Second, we examined the robustness of methods across varying GC content, methylation level, mappability, and characteristics of tumor genome. As a result, methods showed distinct profiles across quantitative dimensions, and conventional quality metrics standard metrics cannot capture quantitative errors. Specifically, non-targeted NGS assays were dominated by sequencing depth-dependent binomial sampling noise. To address this, we developed MethRidgeQC, a reference-free quality control tool that can classify CpG measurements into four reliability classes. Retaining high-reliability classes reduced RMSE by 51–74% compared with low-reliability measurements, outperforming depth-based filtering by up to 47% while preserving 54–79% of CpGs for downstream analyses. Our study provides guidance on method selection, and actionable, reference-free quality control for various non-targeted NGS-based methods.