Abstract:
|
High throughput bisulfite sequencing has been widely used in genetics and genomics, providing accurate measurements of methylation level at every site along the genome. These sequencing studies often employ related individuals, and individual relatedness presents a major computational hurdle to analyzing these count based data sets. Previous approaches either model the count data based on beta binomial models that fail to account for individual relatedness, or model normalized data using linear mixed models that fail to account for the discrete nature of these data sets. To improve on previous methods, we propose a Binomial mixed effects model that not only accounts for individual relatedness but also works on raw count data. We present a novel and efficient sampling-based inference algorithm for the model, taking advantage of the recently described computational improvements for linear mixed models. Using real data sets and real data based simulations, we show that our model is more powerful and better calibrated (i.e. correct type I error control) in related individuals than several widely used approaches in detecting differentially methylated sites.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.