|
Abstract:
|
Often in genomics, scientists search for regions of the genome that are involved in biological processes. High-throughput technologies are used to define regions of interest while screening away most of the genome, as only a handful of regions will be verified in expensive follow up studies. Due to non-stationarity, it is hard to decide which regions to keep based on heuristics or even p-values. In this talk, I will describe our approach to characterizing such regions in terms of an interpretable population parameter. By methods of conditional inference, we can estimate and derive region-wise intervals for this quantity, accounting for both the selection and the local covariance. I will demonstrate the method effectiveness for evaluating differentially methylated regions (mDNA) between different tissues types.
|
