Section 6: What Have We Learned?

The drop in sequencing costs over the last ten years has led to a need for automized analysis pipelines and more computational / storage power to handle the vast flood of data being generated by a multitude of parallel sequencing efforts. Two major tasks of cancer genome projects going forward can be roughly grouped into two areas: characterization and interpretation.

For characterization, there seems to still a need for a systematic benchmark of analysis methods (one example is ROC curves - curves that illustrate the performance of a classifier with a varying discrimination threshhold). We saw that cancer mutation rates tend to vary more than 1,000-fold across different tumor types. We also learned that clonal and subclonal mutations could be used for studying tumor evolution and heterogeneity.

Running a significance analysis on the sequencing results identified a long-tailed distribution of significantly mutated genes. Since we’re dealing with a long tail distribution, we can increase the predictive power of our models and detect more cancer genes by integrating multiple sources of evidence. However we have to take into account that mutation rates differ according to the original sample, gene, and category from each study.

---