Fig. 1

Assembly of the CSN and BSN. The experimental procedure leading to the assembly of the CSN and one of the three BSN (here the BSN-BP) is depicted schematically, in the top and middle panels, respectively. The PS (black circles), labeled A, B and C, are annotated with the HPO terms (blue squares) and the GO-BP terms (red hexagons) labeled 0, 1, 2, 3 and 4. As shown in the schematic ontology trees, 4 is the common ancestor of terms 2 and 3, and 0 is the root term. Then the algorithm retrieves all the possible PS-PS pairs and the annotations that the two PS share (dotted squares and hexagons). The shared annotations can be either an identical term (e.g., A and B sharing 1) or the most informative common ancestor of two different terms (e.g., A and C sharing 4 as a common ancestor of 2 and 3); highly dissimilar PS (e.g., B and C) share the root 0. For simplicity’s sake, each pair of PS is assumed here to share only one annotation. At the end of the search, a non-weighted bipartite graph is assembled. Then, from the PS-HPO and PS-GO bipartite graphs, the CSN and the various BSN are derived, by linking all the PS that share the annotations. The thickness of the edge in the CSN and BSN is proportional to the similarity coefficients of the linked PS pair. The bottom panel shows schematically different types of correlation between HPO- and GO-based similarities. For instance, in the AB, AC and BC pairs of PS, both the clinical and the biological similarities are directly correlated in a high, intermediate or low manner, respectively. In contrast, two additional PS pairs exemplify instances of high clinical (but low biological) similarity (AX) and, conversely, high biological (but low clinical) similarity (AY)