Figure 1

Separation of tumor from blood by Fastbreak (left) and BreakDancer (right). The separation of the tissue type (cancer versus blood), showing kernel density estimates along the first coordinate of a multidimensional scaling (MDS) solution derived from the mutual information distance [7] between samples, demonstrates that Fastbreak is robust against sample collection and instrumentation biases. The blood sample densities are shown in green, while the pooled cancer samples are shown in grey. For Fastbreak (left) the blood samples show a high c-index [8] for the blood cluster (0.97) and no significant correlation (−0.01) between SVs detected in genes and their corresponding coverage; for BreakDancer [6], the blood samples show a lower c-index for the blood cluster (0.68) and a weak correlation (0.28) between SVs detected in genes and their corresponding coverage, an undesired confounding property. Both applications were run only on samples and regions that passed our QA process. Without this restriction, variation along the primary coordinate of MDS is dominated by batch effects unrelated to tissue type.