All samples were obtained under institutional IRB approval and with documented informed consent. A complete list of samples is given in Table S2. Whole-exome capture libraries were constructed and sequenced on Illumina HiSeq flowcells to average coverage of 118x. Whole-genome sequencing was done with the Illumina GA-II or Illumina HiSeq sequencer, achieving an average of ~30X coverage depth. Reads were aligned to the reference human genome build hg19 using an implementation of the Burrows-Wheeler Aligner, and a BAM file was produced for each tumor and normal sample using the Picard pipeline6 (link). The Firehose pipeline was used to manage input and output files and submit analyses for execution. The MuTect30 and Indelocator (Sivachenko, A. et al., manuscript in preparation) algorithms were used to identify somatic single-nucleotide variants (SSNVs) and short somatic insertions and deletions, respectively. Mutation spectra were analyzed using non-negative matrix factorization (NMF). Significantly mutated genes were identified using MutSigCV, which estimates the background mutation rate (BMR) for each gene-patient-category combination based on the observed silent mutations in the gene and noncoding mutations in the surrounding regions. Because in most cases these data are too sparse to obtain accurate estimates, we increased accuracy by pooling data from other genes with similar properties (e.g. replication time, expression level). Significance levels (p-values) were determined by testing whether the observed mutations in a gene significantly exceed the expected counts based on the background model. False Discovery Rates (q-values) were then calculated, and genes with q≤0.1 were reported as significantly mutated. Full methods details are listed in Supplementary Information.