Fig. 4

Error rates exhibit inter- and intra-dataset variability. The violin plots show the distribution of per-sample precision and recall measurements. The top panels show precision and recall at heterozygous sites. The bottom panels show precision and recall at homozygous alternate sites. The grey bars indicate the most extreme precision and recall measurements that can be supported by the SNP density of the dataset. Unsurprisingly, precision and recall vary across sequencing datasets. However, we noticed unexpected sample-level variability in precision and recall within datasets, indicating that samples sequenced on the same platform and analyzed with the same software pipeline may have dramatically different error rates