Fig. 6

The loss of power of the TCGA CNA pipeline in low-purity samples. The copy number of a genomic segment predicted by the TCGA pipeline, which does not model the tumor purity, is a weighted average of its respective copy numbers in the tumor and normal cells. As the purity of a tumor sample declines, the increasing fraction of normal cells will move the predicted average copy number closer to two. Accucopy treats the copy numbers of the tumor and normal cells within one tumor sample as two separate parameters in its model. a Exhibits the copy number profile of a low-purity sample (purity = 0.286) predicted by the TCGA pipeline, the upper panel, versus that predicted by Accucopy, the lower panel. The FullC between the two profiles is 0.523. b is a similar plot to panel A, for a high-purity sample (purity = 0.911). The FullC between the TCGA-pipeline and Accucopy predicted CNA profiles is 0.877. In both samples, the segmentations of the genome by the TCGA pipeline and Accucopy are highly similar. In addition, the copy number qualitative predictions (duplication or deletion) for individual segments are also highly similar. However, in the high-purity sample (b), the copy number quantitative predictions of abnormal segments are further away from two and are numerically more concordant with those by Accucopy, manifested by a higher FullC than the low-purity sample (a)