Supplementary Figure 1: Integration and selection of cancer events in HNSCC.

(a) Tumor data first pass an integration step in which knowledge of pathway or chromosomal structure is used to create meta features. Data are then filtered on the basis of event frequency across tumor samples or comparison with matched normal samples, yielding a pool of candidate cancer-associated events. (b) Example of the integration step in which sparse mutations to the SOS1/RAS pathway (Reactome 524) are combined to derive (c) a single pathway mutation marker for each patient. In b, green bars indicate that a patient (column) has a mutation in a particular pathway gene (row). (d) Example integration of mRNA expression on a pathway, in which principal-component analysis (PCA) is applied to the gene-by-patient expression matrix. Shown are the gene loadings for PCA of the PIP3 signaling pathway (mSigDB M1315). (e) In each patient, the first principal component is used to represent the consensus expression value of the pathway. Here the blue bars represent patients for whom this value is above the threshold and for whom the pathway is scored as ‘upregulated’.