Normalization of Batch Effect in Large Scale DNA Methylation Profiling

Massively parallel sequencing technologies permit the rapid production of large data sets which have the potential to markedly accelerate genetics and biological research. Reduced representation bisulfite sequencing (RRBS) is a powerful and cost-effective tool for investigating DNA methylation at the genomic scale. Further advancement in RRBS and its technology requires effective normalization methods of the sequencing data generated from different batches for better quantitative DNA methylation analysis. Currently, there are no normalization tools for such data. Variations in library preparation methods as well as sequencing runs can produce significant differences in DNA methylation levels that mask the biological significance. We have generated and analyzed 20 RRBS libraries from mice liver, with varying PCR cycle numbers, library clean up methods as well as sequencing run conditions. Based on the observed correlation between methylation levels and the batch factors, we propose a regression based normalization method to normalize the data from different conditions. Our method facilitates the comparisons of DNA methylation profiles across samples generated from different batches and makes bisulfite sequencing better suited for large-scale DNA methylation studies.