Computational methods for cancer driver discovery: A survey

Abstract

Identifying the genes responsible for driving cancer is of critical importance for directing treatment. Accordingly, multiple computational tools have been developed to facilitate this task. Due to the different methods employed by these tools, different data considered by the tools, and the rapidly evolving nature of the field, the selection of an appropriate tool for cancer driver discovery is not straightforward. This survey seeks to provide a comprehensive review of the different computational methods for discovering cancer drivers. We categorise the methods into three groups; methods for single driver identification, methods for driver module identification, and methods for identifying personalised cancer drivers. In addition to providing a "one-stop" reference of these methods, by evaluating and comparing their performance, we also provide readers the information about the different capabilities of the methods in identifying biologically significant cancer drivers. The biologically relevant information identified by these tools can be seen through the enrichment of discovered cancer drivers in GO biological processes and KEGG pathways and through our identification of a small cancer-driver cohort that is capable of stratifying patient survival.

Keywords: cancer driver; cancer driver discovery; coding gene; computational method; microRNA.

Figure 1

Cancer drivers and genes with mutations. Genes with driver mutations are cancer drivers. Some genes which do not contain mutations but regulate driver mutations to develop cancer are also considered as cancer drivers.

Figure 2

Categorisation of cancer driver discovery methods. The methods are categorised in three groups: Single cancer driver identification, Cancer driver module identification, and Personalised cancer driver identification. Single cancer driver identification includes two sub-groups: Mutation-based methods and Network-based methods. Mutation-based methods discover cancer drivers using mutation significance, functional impact of mutations, etc. Most cancer driver module identification methods use the mutual exclusivity of mutations to identify modules of cancer drivers.

Figure 3

Comparison of F₁Score of ActiveDriver, DawnRank, DriverML, DriverNet, MutSigCV, OncodriveFM, PNC, and SCS in identifying coding cancer drivers at the population level. The x-axis indicates the eight methods and the y-axis shows the F₁Score. The results are based on the cancer driver prediction for the five cancer types, including BRCA, LUAD, LUSC, KIRC, and HNSC, of the eight methods.

Figure 4

Overlap among the cancer drivers predicted by different methods. The charts illustrate the overlap among the cancer drivers at the population level predicted by the five methods (DriverML, ActiveDriver, DriverNet, MutSigCV, and OncodriveFM) w.r.t the five cancer types, including BRCA, LUAD, LUSC, KIRC, and HNSC. In each chart, the horizontal bars at the bottom left show the number of detected cancer drivers validated by the CGC, the vertical bars and the dotted lines show the overlap of the validated cancer drivers of the methods. If there is not an overlap, it will be a black dot.

Figure 5

Survival curves, clustering display, and silhouette plot. Survival curves are for cancer subtypes identified by using the four predicted cancer drivers, including AKT1, PTEN, CDKN1B, and TP53. The survival curves show the significant difference in the survivals of patients of the two subtypes (p-value = 0.0245). The clustering display indicates a highly qualified clustering with the similarity of samples in each subtype (i.e. Light dots show the similarity of samples). The silhouette plot has a large average silhouette width (0.76/1), indicating the clustering validity when using these four genes.