Abstract:
|
Recently Bayesian methods have been widely used in genomic data analysis. Despite that many methods have been developed in the Bayesian literature, the computation of the marginal likelihood, which plays an important role in model comparison and averaging in Bayesian inference, remains a challenging problem particularly in high-dimensional settings, which are often encountered in genomics. We develop a new approach, MCMC-CE, which involves drawing samples from the posterior distribution via Markov chain Monte Carlo (MCMC) and using them to estimate the marginal likelihood with the cross-entropy (CE) method, for efficient and accurate computation of the marginal likelihood. By applying MCMC-CE to simulated and real-world datasets, we show that this approach not only outperforms several widely used existing methods on accuracy and computational speed, but also has a broad range of applications in genomic data analysis.
|