Network Analysis of Mutations Across Cancer Types. M.DM. Leiserson1, F. Vandin1, H.-T. Wu1, J. Dobson1,2, A. Gonzalez-Perez3, D. Tamborero3, N. Lopez-Bigas3, B. Raphael1 1) Department of Computer Science, and Center for Computational Molecular Biology, Brown University, Providence, RI; 2) Department of Molecular Biology, Cell Biology & Biochemistry, Brown University, Providence, RI; 3) Experimental and Health Science Department, University Pompeu Fabra, Barcelona, Spain.
Recent large-scale cancer sequencing studies have revealed extensive mutational heterogeneity across samples from the same cancer type with relatively few genes recurrently mutated across many samples and a long tail of many genes, each mutated in a small number of samples. A major reason for this heterogeneity is that cancer mutations target cellular signaling and regulatory pathways, and different combinations of mutations may perturb these pathways in different individuals. This complicates efforts to distinguish driver from passenger mutations according to their observed frequencies.
We introduce a new algorithm, HotNet2, to identify subnetworks of a protein-protein (or protein-DNA) interaction network that are mutated in a statistically significant number of samples. HotNet2 uses an insulated heat diffusion model to simultaneously assess both the significance of mutations in individual proteins and the local topology of a proteins interactions. Thus, HotNet2 is not restricted to analysis of known gene sets, but can discover novel combinations of interacting genes. We score the significance of individual proteins using: (1) the frequency of mutation; (2) the accumulation of mutations with high functional impact across tumor samples computed by Oncodrive-FM (Gonzalez-Perez and Lopez-Bigas 2012); (3) bias towards the gene misregulation due to copy number aberrations computed by Oncodrive-CIS (Tamborero et al. 2013).
We used HotNet2 to perform a pan-cancer analysis of mutated networks using whole-exome sequencing and copy number aberration data from 2866 samples from The Cancer Genome Atlas (TCGA) across twelve different cancer types. We used two networks: one consisting of high quality protein-protein interactions (Das and Yu 2012) and a second consisting of a variety of interaction types (Khurana et al. 2013). We identified >10 significantly mutated subnetworks (P < 0.01) in each case. These subnetworks overlap well-known cancer signaling pathways (e.g. p53, RTK, and RB), but also include subnetworks with less characterized roles in cancer; e.g. the cohesin and condensin complexes and the SLIT-ROBO pathway, the latter involved in cell migration. Several of these subnetworks are significantly enriched for mutations in a specific cancer type (e.g. chromatin related genes including ARID1A and PBRM1 in renal cell carcinoma).
You may contact the first author (during and after the meeting) at