[Wishlist] DMR Detection Methodology

[nhejazi]

We should support the finding of differentially methylated regions (DMRs) in two ways:

1. Design of a nonparametric method for assessing whether a region may be labeled as a DMR --- this should build upon work in the detection of sparse signals, using the test statistic for the TMLE of the target parameter of choice as a score, over which such techniques may operate (e.g., see the work of T. Cai, X. Lin). This idea stemmed from helpful conversations with Rajarshi Mukherjee.

2. Integration with already existing techniques/software for the detection of DMRs. Algorithms and software like bumphunter generate p-values from test statistics derived from parametric models; however, the DMR detection methods themselves are nonparametric usually (e.g., permutation tests). It should be possible --- even relatively trivial --- to use the p-values derived from the TMLE procedures provided as input to the DMR-finding algorithms of other Bioconductor packages. This would amount to a very useful integration with existing software. This idea stemmed from helpful conversations with Alan Hubbard.