Correction

After my initial submission of this decision letter, I realized I had made a mistake in items 7 and 8 below. I correct this mistake in this opening section and then continue with the decision letter as originally submitted, mistakes and all. The previous version claimed that the genotype likelihood is binomial. But because error ( ) varies among sequencing reads, the probability (p) of observing the reference allele also varies, and the likelihood is not binomial. To calculate it without approximations, one would need to sum across all ways of partitioning the C reads among the two alleles of each heterozygous genotype. Avoiding this sum requires approximations, even in the diploid case. For example, consider the model of Li et al. [2008, sec 1 and Eqns. 9–11 of Supplementary Materials]. Their approach is similar to that of the current manuscript in that it estimates each genotype from sequencing reads at an individual nucleotide site, rather than from several linked sites. It differs in that it deals only with diploids. To avoid summing across partitions, those authors approximate the likelihood of heterozygote genotypes using a binomial formula that ignores sequencing error altogether. In the manuscript of Sorragi et al, the central problem is a lack of clarity in section 1.2 of Supplementary Materials, both in the text and in the equations. In addition to the points I make below, I would add that we need some discussion of the approximations used to avoid the sum over partitions.