Beta binomial logistic regression with empirical Bayes shrinakge of dispersion estimates for differential RNA editing analysis

RNA editing is a post-transcriptional modification, namely an alteration in the RNA molecule sequence in which adenosine and cytosine bases are substituted with inosine and uracil bases. This process is carried out by enzymes called ADAR and APOBEC and found important for normal neuronal function and the immune system. Moreover, these modifications were found to be implicated in several medical conditions.

Even though this modification is well-studied, there was no broadly accepted statistical model to test for differences in RNA editing levels. Recently, the beta-binomial distribution was demonstrated to fit well to count data generated by RNA sequencing experiments in which a portion of reads covering a specific segment in the genome are edited, i.e., displayed an alteration in the RNA sequence.

A major component in statistical inference is to accurately estimate the dispersion of the observed data. DREDD efficiently estimates the dispersion parameter of the beta-binomial likelihood function from RNA editing count data. Then, by modeling the mean-dispersion trend, in a similar approach implemented in DESeq2 for differential expression analysis, DREDD provides a correction to the dispresion paramter or more accurate differential editing results.