The emergence of cancer genomics, combined with increased understanding of the molecular basis of oncogenesis, has stimulated hope that treatment will improve by becoming more targeted and individualized in nature. Cancer genomics studies established a number of critical cancer genes, leading to a number of successful targeted therapies (e.g. Gleevec, Herceptin and Plexxikon). Despite these successes, most cancers do not have a targeted therapy and when one exists, response is highly variable, even among patients that share the targeted mutation and tumor type. To move cancer into the era of personalized therapies, it becomes important to identify the alterations driving tumor progression in each tumor, determine the network that links these aberrations, and identify factors that predict sensitivity to targeted therapies. As projects such as The Cancer Genome Atlas (TCGA) amass cancer cell genomes at a breathtaking pace, a staggering genetic complexity is revealed. To interpret cancer genomes, a key computational challenge is to separate the wheat from the chaff and define both what are the key alterations likely to be functionally driving cancer and then, after defining such genes, begin to identify mechanisms of action and therapeutic implications. Leveraging components from our published methods, CONEXIC (Akavia et.al Cell 2010) and LirNet (Lee et.al, PLOS Gen 2009), we will develop machine-learning algorithms that integrate cancer genomic data to do just that. We will apply the methods we develop to melanoma, glioblastoma, ovarian, breast and colon cancer and experimentally follow up on our computational findings, towards a better understanding of each of these deadly cancers. The approaches developed in this grant will accelerate discovery to rapidly extract the maximal value from modern genomic studies and help carry cancer genomics from the diagnostic to the therapeutic realm.