Fig 1.
Details of the computational framework of MEDICI.
Curated pathway descriptions are integrated with novel interactions discovered by PPI screening to generate an interaction superpathway. Gene essentiality measurements are layered onto the nodes of the superpathway, and the network topology is transformed to the dual graph where the genes become network edges and the gene-interactions become network nodes. Gene essentialities are then diffused over their interactions to infer interaction essentiality weights.
Fig 2.
Clustering of most essential PPIs in the superpathway.
(A) Unsupervised Hierarchical Clustering of the 360 most essential PPIs across the 165 cell lines identifies 12 major clusters. The 360 PPIs with an average essentiality score > 0.5 were used to cluster 165 cell lines used in the Achilles shRNA screening study using Cluster and Java Treeview software [29]. PPI essentiality data was median centered and clustered by average correlation. Red indicates higher essentiality and blue indicates lower essentiality. Major hubs for each cluster are indicated on the right. (B) Clustering of 5798 PPI-MPER values across 165 cell lines. Red indicates PPI essentiality is greater than the max protein essentiality, and blue indicates the PPI essentiality is less than the max protein essentiality.
Fig 3.
Correlating interaction essentialities with drug sensitivity measures provides insights into mechanisms of action.
We correlated drug sensitivity measures from CCLE with interaction essentiality scores to identify critical interactions that predict therapeutic sensitivity. Sensitivity to the MAPK inhibitor AZD6244 is highly correlated with PRKDC-TP53 interaction essentiality, which is consistent with the well established role of p38-MAPK in cell cycle arrest in response to DNA damage [51–53].
Table 1.
Correlations of PPI essentiality with drug sensitivities for 19 CCLE compounds with PPI essentiality data for their respective targets.
Fig 4.
PPI networks associated with genetic mutations.
(A) Networks of PPIs most increased in essentiality in cells with mutation or loss of the PTEN tumor suppressor gene. The 14 most significant PPIs are shown. Significant differences in PPI essentiality were computed in GenePattern [54]. Networks were visualized with Cytoscape [55]. (B) Networks of PPIs most increased in essentiality in cell lines with mutation or loss of the APC tumor suppressor gene. The 20 most significant PPIs are shown.
Fig 5.
PPI Essentiality association with patient survival.
(A) QQ plot of observed vs. expected log-rank p-values for LUAD patients split based on 5798 PPIs. (B) Network of PPIs with significant log-rank p-value for discriminating survival for TCGA LUAD patients that is centered on JAK1. (C) Kaplan-Meier curve of TCGA LUAD patients separated based on the presence or absence of the JAK1-PIK3R1 PPI. Patients without the JAK1-PIK3R1 PPI have improved survival compared to patients who retain this PPI.