Comparison of the results of the three settings of the BeWith schema on real and randomized UCEC data.

We also performed additional evaluation of our method by running BeWith on simulated data to validate if BeWith indeed find modules with the desired properties in presence of random noise (See Section C in S1 Text for discussion). The results demonstrated that our method is robust and recover the planted modules in most instances (≥99%). The accuracy decreased if the planted module was increasingly noisy (See Table A in S1 Text). In addition, we compared our modules with the modules identified by other module detection algorithms and showed that our method is better than or comparable to previous methods in terms of cancer driver identification, while our modules have additional properties such as co-occurrences and functional coherence (See Section D in S1 Text).