Data normalization for DNA methylation profiles of cancer genomes

For DNA methylation data, many researchers believe that current normalization methods may distort the real difference between the methylation signals for normal and cancer samples. In this paper, using six methylation datasets for three cancer types, we found almost all of the differentially methylated (DM) genes detected in both the Quantile-normalized data and non-normalized data show consistent methylation states. In each dataset, more DM genes could be found in the Quantile-normalized data than in the non-normalized data. We showed that, for each dataset, the DM genes additionally selected in the Quantile-normalized data showed significantly consistent methylation states in another independent dataset for the same cancer, indicating that these extra DM genes are effective biological signals. Hence, the Quantile normalization method can be applied to normalize methylation data to increase the power of detecting DM genes. However, in some datasets, the methylation states of about 1% of DM genes selected in the Quantile-normalized data are opposite to those in the non-normalized data and we must filter out these false DM genes.