Fig. 6
From: MUFFINN: cancer gene discovery via network analysis of somatic mutation data
MUFFINN is not significantly improved by using MutsigCV scores. MUFFINN was performed with raw mutation frequencies or MutsigCV scores, which are normalized by BMR and mutational heterogeneity. HumanNet was used as the network model. Comparison of predictions for cancer genes annotated by a CGC and b the 20/20 rule. AUC scores for 17 cancer types (LAML was excluded in the analysis due to the indiscriminative MutsigCV scores due to the low mutation rate in leukemia) are represented as box plots. Performance comparisons were also conducted based on the number of retrieved cancer genes annotated by c CGC and d the 20/20 rule in the top 100, 500, and 1000 candidates