Suppose now I've identified putative structural variants (for example deletion) based on discordant read pair and split-read approach. And I would like to further look at the read-depth information within these callings, for both additional supporting evidences (for example, I would toss out the putative callings if read density is quite high within candidate callings of deletion), and genotyping (to know it's heterozygous or homozygous deletion).

I know basically we need to calculate the normalized average read depths within the candidate callings. But I'm wondering what model/statistics I could use? Say, Gaussian mixture model? Any other ideas?

Thanks