Modeling of replicates variances for detecting RNA methylation site in MERIP-SEQ data

The recent advent of the state-of-art high throughput sequencing technology, known as Methylated RNA immunoprecipitation (IP) sequencing (MeRIP-Seq), provided the biologists the first global view of epigenetic modifications on the transcriptome at a high resolution. However, novel and more sophisticated statistical computational methods are needed to detect methylation sites from MeRIP-Seq. Here, we propose a mixture of Beta-binomial model for mathematically modeling the data variance of replicates in the MeRIP-Seq. An Expectation-Maximization algorithm is derived to learn the model parameters and perform site detection. To illustrate the utility of our model, it is evaluated on simulated datasets and a real MeRIP-Seq data for N6-Methyladenosine (m⁶A) methylation. The results show that the model has a higher sensitivity and specificity under various variance conditions than previous methods, demonstrating its robustness on the MeRIP-Seq data.