Figure 2

Kernel density estimates of cancer type samples (glioblastoma versus ovarian cancer) along the first coordinate of a multidimensional scaling solution derived from the mutual information distance between samples as analyzed by Fastbreak (left) and BreakDancer (right). The GBM cancer patient distribution is shown in red (with a c-index of 0.8 for Fastbreak and 0.94 for BreakDancer [8]), and the ovarian distribution (with a c-index of 0.7 for Fastbreak and 0.95 BreakDancer) in blue. The results show that Fastbreak can distinguish cancer types without exhibiting strong batch effects. Some of BreakDancer’s separation of samples can be attributed to batch effects, as shown in Figure 1, due to differential coverage between the two cancers.