For our problem, a Bayes procedure with good frequency properties is particularly attractive. In copy number variation, we are principally interested in excursions into and out of the normal diploid state (or haploid for sex chromosomes) and, it is therefore natural to express interest in a frequentist property, such as the false positive rate, which tells us the long-run frequency with which we would make incorrect CNV detections. In our model, the rate of excursions (and hence our false positive error rate) is controlled by a characteristic length parameter L and a threshold value BFthresh. The greater the characteristic length, the less likely we are to make excursions into and out of the null state, and hence fewer copy number variant events will be called. Furthermore, if an excursion is made, the rate at which we accept this copy number variant is further determined by the significance we attribute to it: we only accept a copy number variant if the Bayes Factor associated with the event is greater than a threshold value BFthresh.
