I have a project where I have developped a predictive model based on SNPs. A colleague of mine developped from its side a model built on the allele coverage where each base pair was allocated a number of mapped reads. I've read that this type of encoding can detect deletions or insertions also large structural variations but I still miss the intuition. Since I'm not familiar with this approach I would like to understand the advantages and inconvenients of such allele coverage encoding for detecting more mutations, hence use them into predictive models.