Fig. 3

Accucopy performance on the HCC1187 dataset. The sequencing coverage for all samples is 10X. The tumor purity varies from 0.1 to 0.9. a The TCN FullC results show the Accucopy TCN calls are at least 90% concordant with the ground truths. The TCN CallF results indicate Accucopy assigns copy numbers to close to 100% of the genome. b The MACN FullC results indicate the Accucopy MACN calls are close to 95% concordant with the ground truth. The MACN inference is better than that of the simulation study under similar conditions because all the true MACNs of HCC1187 are effectively LOHs (Loss-Of-Heterozygosity). The non-LOHs of HCC1187 have unknown MACN state and are excluded in comparison. The statistical power to infer MACNs is higher for LOHs than non-LOHs because the difference between the major and the minor allele copy number is bigger for LOHs. c The top panel is the TCN ground truth based on the spectral karyotyping (SKY) result [25]. The bottom panel are the TCN estimates by Accucopy for a purity = 0.3 sample. The green segments are considered subclonal and assigned with non-integer (i.e. 3.8) copy numbers. d The top panel is the MACN ground truth. The bottom panel are the MACN estimates by Accucopy for the same tumor sample. The Accucopy MACN estimates are only for clonal regions. Note that Accucopy makes MACN estimates for regions where the SKY MACN calls are missing because SKY can only call MACN for LOH regions