Finding signals in Infinium DNA methylation data

2017-07-29DNA methylation plays critical roles in a variety of biological processes, such as aging, carcinogenesis, etc. DNA methylation data from Illumina Infinium platforms are measured as the proportion of total intensities due to methylated alleles. They are values bounded between 0 and 1, assuming to follow Beta distribution. Due to the specialties of this distribution, methods work for Gaussian distributed data, such as gene expression microarrays, may not perform equally well on DNA methylation data. ❧ This dissertation addresses two issues we encounter when dealing with high throughput DNA methylation data from the Infinium platforms. One is to do feature selection for cluster analysis to discover novel sample subgroups, and the other is to correct Batch effect in data that was due to samples being processed and measured in different batches. Methods used in these two processes in order to strengthen the biological signal of interest are evaluated and compared using both simulation and real data sets