Fig 1.
Patient specific GBM tumor mutation profiles were obtained from TCGA. The spatial proximity of each mutation is searched and mutation patches were obtained. Simultaneously, each cancer related driver protein having at least one mutation in each patient was used to reconstruct patient-specific sub-networks. Red dots and stars in the middle panel correspond to mutations mapped to the sequence, structure and PPI network. Finally, the sub-networks were used to classify the patients, to find signature patches in each patient groups and to demonstrate the help of 3D patches in overcoming heterogeneity. Lastly, we investigated the patient groups to find an association with the clinical outcome by using cell line drug sensitivity data. The brain and human icons in the first panel are retrieved from Reactome Icon Library [32].
Fig 2.
The 3D patch profile and survival curve of the patients having at least one patch mutation.
(A) Grouping patients based on 3D patches. Each column represents a patient and each row represents patches that are present in at least 2% of patients. (B) Kaplan-Meier survival curves of the patient groups. (C) Mapping patches of frequently mutated hub proteins to their domains. Red colors represent the presence of patches in the corresponding domains.
Fig 3.
(A) Histogram of the patch sizes for intra- and inter- patches. While frequently mutated hub proteins have relatively large patches, most patches are small in size. (B) PTEN Patch 1—patch size—43 residues—an example of mutations forming a residue network from the surface to the core. (C) EGFR—three patches with different sizes: 11, 14 and 5 residues, respectively. (D) PIK3R1-PIK3CA—three inter-patches—each patch has at least one mutation in each partner protein.
Table 1.
Number of mutations mapped to protein structural regions.
Fig 4.
The characteristics of the mutations.
(A) Changes in the chemical properties of the driver protein mutations according to their physical locations (B) Distribution of mutations according to their disease association (EVmutation score) in different locations. The more negative EVmutation score implies the more damaging mutation. (C) Fraction of mutations according to their disease association (PolyPhen-2 status: benign, possibly damaging, probably damaging) in different locations. (D) Fraction of patch mutations and singletons according to their locations.
Table 2.
Disease association of singleton and patch mutations in the interface region of the hubs and the rest.
Fig 5.
Representation of the proteins that have multiple interfaces used by different subsets of partners and one interface shared by multiple partners.
(A) The left part represents an example for proteins having multiple interfaces (PTPN11) with different subset of partners and the right part represents an example for proteins having one interface (TP53) shared by different partners. (B) The organization of RB1 and PIK3CA interface mutations are represented as a network where the nodes are mutations and proteins and the edges between mutations represents at least one shared partners between two mutations. The edges between proteins and mutations represents the interface (C) The table for numbers of mutations in each interface type.
Fig 6.
Proteins having mutations in their interfaces and the spatial arrangement of the mutations.
(A) The degree distribution of the proteins in the interactome having at least one mutation in the interface region are classified based on the 3D patch organization. Three classes, namely, proteins having multiple patches, single patch and without any patch but only singleton are on the x-axis and log10 of the degree of each protein is on the y-axis. High degree proteins (hubs) in interactome have tendency to include multiple patches on their interface regions. (B) The hub proteins have more interface mutations locating inside the patches. Only the proteins having at least one patch on their interface regions are shown.
Fig 7.
Clustering the tumors based on pathway similarities.
(A) Consensus clustering of the network inferred disease signatures. Each entry in the matrix shows the co-occurence of each pair of patients. (B) Kaplan-Meier survival plots of the patient groups. Each curve represents one group. (C) Enrichment of KEGG pathways across the patient groups. Reds indicate that KEGG pathways are mainly enriched in patients of specific groups except Group 4 which does not have any KEGG pathway dominantly enriched in its patients.
Fig 8.
Linking the network-guided patient groups to drug treatments through 3D spatial patches.
(A) Nodes represent group identifiers, patch names, cell line names, drugs and their targets, respectively. Edges between group Ids and patch names imply the presence of the presence of the corresponding patch in the group. Edges between patch names and cell line Ids represent that at least one mutation in the corresponding patch is present in the connected cell line. Edges between cell lines and drug names imply that the cell line is treated with the corresponding drug. If the cell line is sensitive to the drug then the edge color is red, if resistant the edge color is blue. Edges between drugs and target proteins are to show that proteins targeted by the corresponding drugs that are significantly present in the linked group. (B,C,D) Three submodules are retrieved from the network to show some therapeutic hypotheses where the first one is Pazopanib (targeting CSF1R and PDGFRB) for Group 5, the second is the possible resistance of Group 2 to ATM inhibition and the last one is WZ3105 (targeting SRC) for Group 5.