Abstract:
|
We develop a bivariate Bayesian hidden Markov model (HMM) and an expectation- maximization method for detecting allele-specific expression (ASE) in the control group and alternations of ASE in the case group simultaneously. We name our method the hmmASE algorithm. The hmmASE algorithm gains advantages over existing methods by addressing the limitations of current practice and recognizing the complexity of biology. Specifically, the bivariate Bayesian HMM detects ASE at the exon level rather than testing ASE for a whole gene. In addition, it models spatial correlations among genomic locations, unlike existing methods that often assume independence across observations. The bivariate HMM draws inferences simultaneously for control and case samples that exploits information among observations at the same genomic locus. We use simulation studies that mimic real datasets to evaluate our hmmASE method and other popular approaches. At last, we use a large offspring syndrome study to illustrate the practical utility of our approach.
|