Fig. 5: Evaluation and comparison of the single- and multi-sample intracellular signalling network inference on CPTAC.

a, Overview of the approach. Using transcriptomics from patients, we estimated TF activities with Decoupler, which was used as an input for single-sample and multi-sample CARNIVAL. In both cases, we assume that the inputs (receptors) are unknown, and they are automatically selected during optimization (inverse CARNIVAL). For validation, we used phosphoproteomics data for the same patients, from which kinase activities were estimated. b, Results for the single- and multi-sample CARNIVAL, showing the average number of selected edges from the PKN for all patients (top left), same for the vertices (top right), the intersection of edges across patients (bottom left) and the mean proportion of correctly inferred interactions involving deregulated kinases, used as a validation set. Error bands indicate ±1σ over ten independent runs (with different random seeds) at each value of λ. c, Interactions that appear in the intersection of the inferred networks for the optimal λ = 0.8. To account for variability, the multi-sample inference was repeated 30 times with different seeds, and we kept the interactions of the intersection if they appeared in at least 50% of the alternative solutions. The colours correspond to the average TF activity across patients.