Table 1.
Summary of methods for identifying single cancer drivers
| Method | Description and reference | Additional information |
|---|---|---|
| Single cancer driver identification | ||
| Mutation-based methods (using mutation significance) | ||
| MutSigCV | Assesses the significance of mutations in DNA sequencing to discover cancer driver genes 13 | The result includes false positives (i.e. passenger mutations with a high degree) |
| Mutation-based methods (using functional impact of mutations) | ||
| OncodriveFM | Uses the functional impact of mutations of genes to detect cancer drivers with the hypothesis that any bias of variations with a significantly functional impact in genes can be used to identify candidate driver genes 14 | It can identify driver genes with low mutation recurrence |
| OncodriveFML | Uses the functional impact of gene mutations to reveal both coding and non-coding drivers 23 | It is applied to 19 cancer datasets and detects several well-known drivers |
| DriverML | Uses the functional impact of mutations to unravel cancer drivers through a supervised machine learning approach 24 | It can be improved if integrating additional well-annotated datasets (e.g. CGC) into the training data |
| Mutation-based methods (using structural consequences of gene mutations) | ||
| ActiveDriver | Looks at the enrichment of mutations in externally defined regions to uncover cancer driver genes 25 | It only analyses missense mutations while other mutations are also important such as in frame del, frame shift del, etc. |
| SGDriver | Uses a Bayes inference statistical framework to incorporate somatic missense mutations into protein-ligand binding-site residues in order to figure out the functional role of the mutations 26 | It can be improved if integrating more mutation types and using molecular network to identify the interacting partners of mutated proteins to expand the candidate pool |
| AlloDriver | Maps mutations to allosteric/orthosteric sites derived from the three-dimensional protein structures to detect potentially functional genes/proteins in cancer patients 27 | It also uses only missense mutations |
| OncodriveCLUST | Detects cancer genes with a large bias in clustering mutations based on the idea that gain-of-function mutations usually cluster in particular protein sections and these mutations contribute to the development of cancer cells 15 | It cannot identify cancer drivers whose mutations are distributed across the sequence |
| Mutation-based methods (others: combining with gene expression, pathways, protein structures) | ||
| IntOGen-mutations | Uses somatic mutations, gene expression, and tumour pathways to identify cancer drivers for various tumour types by combining OncodriveFM and OncodriveCLUST 28 | It can discover driver mutations which are distributed across the sequence and have significant functional impacts |
| PathScan | Combines genomic mutations with the information of genes in known pathways to uncover cancer driver genes 29 | It can be extended to integrate other types of genetic anomalies |
| Sakoparnig et al. | Introduces a computational method to detect genomic alterations with low occurrence frequencies based on mutation timing 30 | It may not discover drivers which are already present at very early cancer stages as we cannot observe a steep rise for them |
| CONEXIC | Applies a score-guided search to detect combinations of modulators which reflect the expression of a gene module in a set of tumour samples then it identifies those which have the highest score in amplified or deleted regions 31 | It is mainly bases on copy number aberrations |
| ncDriver | Screens non-coding mutations with conservations and cancer specificity to reveal non-coding cancer drivers 32 | It tests both recurrence and distribution of mutations to identify cancer drivers |
| HotSpot3D | Identifies spatial hotspots to interpret the function of mutations in the encoded protein 36 | It can detect rare cancer drivers |
| 3D clusters | Clusters somatic mutations in cancer to identify rare mutations based on 3D protein structures 37 | It is limited due to the lack of complete protein structure data for several genes |
| Network-based methods | ||
| Vinayagam et al. | Applies controllability analysis on the directed network of human protein-protein interaction to identify disease genes 38 | As it uses a general protein network (i.e. not specific for a cancer type), uncovered drivers are not particular for any cancer type |
| CBNA | Identifies coding and miRNA cancer drivers by analysing the controllability of the miRNA-TF-mRNA network and mutation data 18 | It builds the gene network for a specific cancer type, thus the results are for the cancer type of interest |
| DriverNet | Uncovers cancer drivers by evaluating the influence of mutations on transcriptional networks in cancer 16 | It relies on a predetermined influence graph which is sparse and incomplete |