Fig. 2

Performance of the network-based method and its components. The performance (i.e., balanced accuracy) of classification models in various settings, and for the three clinical endpoints of interest. a Performance of classical, topological and integrated models on the large cohort (498 samples). b Performance of classical, topological and integrated models on the small cohort (142 samples). c Performance of models using only one of the four feature sets at once (Centrality, node2vec, diffusion and modularity) or all of them (topological, as in a). Results were obtained on the large cohort. d Performance of models using a single centrality metric or all centrality metrics at once. Results were obtained on the large cohort