|
Abstract:
|
Illumina's HumanMethylation450 platform has been very popular for high-throughput DNA methylation analysis. However, technical artifacts are a concern for data processing, including background fluorescence, dye-bias, bias caused by type I/II probe design, and batch effects. Several approaches and pipelines have been developed to address these biases. We evaluate the effect of combining separate approaches to improve signal processing. In our study nine processing methods were applied and compared in four datasets. The comparisons showed an advantage to using BMIQ to adjust type I/II bias in combination with Noob, a within-array procedure to correct background signals and dye-bias, or with Funnorm, a between-array procedure utilizing control probes. Within-array methods were preferred for samples with global methylation alterations between cases and controls. When the batch effects explained more variation, however, the addition of between-array method Funnorm showed a modest advantage, and the additional use of RUVm, a new batch correction method, noticeably improved reproducibility of differential methylation results.
|
