The manuscript describes a novel approach to predict tumor-driving signaling pathway activity. Results are presented in a clear way and the paper is generally well-written. Enough details are presented (mostly in the supplementary information) to be able to replicate the study, which is important because the results are quite convincing. The discussion section is extensive and really puts this study into perspective.

I do, however, have some minor remarks: - The research question could be outlined better. I would say this would be something like 'Does our computational model indeed predict tumor pathway activity better than current methods?' - I wonder about the table describing the probabilistic relationships between transcription complex and target genes, on the left bottom of page 2938. Why are the 'present' vales for upregulated target genes 0.30 and 0.70, while the 'absent' values are 0.45 and 0.55 for downregulated target genes? Shouldn't these be the same, and why where these values 'manually set' anyway? The last sentence of the 'Target gene selection' paragraph (page 2938) states 'These numbers of genes are on the one hand low enough to give specific results, but large enough to get robust models'. How do you know if the models are robust? How do you measure this, and what is the threshold here? - It might be better to separate figure 5 into two figures: one for Wnt (A-B) and one for ER (C-F), to clearly separate the results from the two pathways. - Clearer titles of the figures would have been nice (perhaps not supported by the journal's layout).