Fig. 6

Multi-omics decreases the false positive rate. A Balanced accuracy of random forest classifiers trained on each of the tumor types. The classifiers have excellent performance on the same samples they were trained on or on similar tumors (such as the Kildisiute et al. neuroblastoma sample), while they fail to generalize to other tumor types. B Training on multiple tumor types does not improve the generalization of the classifiers. C Cells misclassified by ikarus can be discriminated based on the average CNV and variance of CNV values. Tumor cells misclassified as normal cells have significantly higher values of both the average CNV and the variance of CNV, than the corresponding normal cells. D Integration of the CNV proofreading decreases the false positive rate from 4% to 1%, with the same average balanced accuracy