Figure 1.

Percentile-percentile (PP) plot comparing the performance of MixMir with five different values of the kmer length k, which defines the way in which the relationship matrix was constructed. The observed P-values are found on the y axis and the expected P-values are found on the x axis. When P-values are correctly estimated, we would expect the observed and expected P-values to be similar, thus approaching the y = x line. Here LM Bin represents the categorical linear model, and MixMirk represents results from MixMir with kinship matrix estimated using kmers. We found that higher values of k were better at correcting for skewness in the PP plots (i.e. false-positive predictions).