The Analysis of Differential Methylation Hybridization Data
The Huang lab has been a pioneer in high-throughput global CpG island methylation detection: The differential methylation hybridization (DMH) methodology allows one to simultaneously observe methylation signatures of thousands of DNA CpG islands. As with all microarray data, the power of the global assays can be greatly reduced by the large amounts of non-biologically relevant signal (a.k.a noise). In an approach to remove signal from the data related to probe composition (i.e., nucleotide related hybridization bias) and DNA characteristics (i.e., over or under abundance of methyl-sensitive restriction cut sites), we have developed a regression model approach to signal preprocessing tailored to the DMH experimental protocol. Concurrently, we are developing a hidden Markov model (HMM) for analyzing the differential methylation signature: The methyl-status of the CpG dinucleotides are modeled as the hidden signal and the probe intensities as the observed. In this manner we are able to assign statistical significance to the methylation status of a given DNA region. As a means of testing both the preprocessing methodology as well and the HMM signature detection technique, we have developed data simulation strategies that model all aspects of the DMH protocol as well as the DNA-probe interactions.