Personalized prioritization of cancer driver mutations

Introduction:

During cancer evolution, cells accumulate DNA alterations that disrupt their natural mechanisms, causing abnormal proliferation that develops into tumors. These alterations are known as "driver" mutations. In contrast, "passenger" mutations are acquired extensively in cancer cells as a byproduct of the cancerous process but have no impact on its progression. Distinguishing driver from passenger mutations and assessing their impact on the individual patient are primary goals in cancer therapy: Knowledge of the pathways in which a mutated driver gene operates can illuminate the disease mechanisms and indicate potential therapies and drug targets. Many algorithms were developed to identify driver mutations in a cohort of patients and some knowledge-bases for experimentally validated driver mutations were constructed, but driver mutation identification in individual patients remains a challenge.

Methods:

We developed a new algorithm called PRODIGY for personalized prioritization of driver mutations by analyzing the entire expression and single nucleotide variation profiles of an individual and incorporating prior knowledge on curated pathways and a large protein-protein interaction (PPI) network. Our algorithm quantifies the impact of each observed mutation on a pathway by delineating paths from the mutation to differentially expressed genes that participate in that pathway, using both pathway and network interactions. Mutations are ranked by their aggregated impact over all pathways.

Results:

In testing colon, breast and lung cancer cohorts and comparison to validated driver genes, PRODIGY outperforms extant prioritization methods. Interestingly, non-personalized prioritization based on common centrality measures of the PPI network outperforms all other extant methods. Our results show that PRODIGY is capable of identifying even rare drivers observed in a few individuals.

Conclusion:

PRODIGY is a novel, powerful algorithm for personalized prioritization of driver mutations that outperforms extant methods. The algorithm can be used by oncologists as a decision support tool for personalized care by choosing the treatment in view of the highly ranked mutations.