Figure 3

Switch errors and quality metrics. We present four scenarios involving low- and high-frequency switch errors and the resulting values for two quality metrics: total switch rate, and the pairwise metric 2_*cse-1, which we term the `phase accuracy (Box 3). The latter was computed on the assumption that each local `back and forth switch error affects 1% of the markers in the chromosome. The true haplotypes (maternal (M) and paternal (P)) are shown at the top. In Example 1, there is a single phase-error switch at `a; variants on opposite sides of the error are incorrectly phased (denoted by the `x on the curved arrow). This situation might arise if a parents haplotype is inferred by subtraction of a childs haplotype, missing a meiotic recombination. Example 1 thus has a low-frequency phase error but no high-frequency errors. In Example 2, there are many high-frequency errors, but no low-frequency errors. Most variant pairs are separated by even numbers of switch errors, and are thus properly phased (curved arrows). This error pattern might arise from a long-read technology such as strobe sequencing. In Example 3, there are three switch errors, one at `a and two at `b. This haplotype could have arisen from a false haplotype assembly join at `a and a sequence error of a single base at `b. Example 3 thus has a mix of low- and high-frequency errors. In terms of the pairwise metric, the single switch in Example 1 most strongly affects the long-range haplotype quality, while the several localized switch errors (each pair affecting just one or a few markers) degrade haplotype quality only modestly.