Fig. 6.

Performance assessment by spike-in experiment. CNV signals are in silico spiked in with increasing population frequencies. a Ground truth and normalization results by CODEX, CODEX2, and SVD-based method. b Histogram of normalized z-scores as $log\left(\mathit{Y}/\widehat{\mathit{\lambda }}\right)$ in CNV (red) and diploid (gray) regions. The z-scores returned by XHMM are on a different scale with a much wider range and significant overlap between the null and non-null. c Segmentation results where CODEX2 has the highest concordance with the ground truth among the methods being compared