Fig. 2

Model performance for different inputs. DNN models relying on different feature sets are compared by reporting their performance on the validation data for ‘Death from disease’ (a) and ‘Disease progression’ (b). Feature sets are defined by the original data that were used (microarray data, RNA-seq data or the integration of both) and by the topological features considered (centrality, modularity or both). Each single point represents a model. For each feature set, several models are trained by varying the neural network architecture and by performing replicates