Table 3.

A set of prediction tools specialized to predicting the disease-driver status of variants in the human cancer genome. Although the generic predictors of Tables 1 and 2 have been used successfully for variant prediction in the cancer genome, more specialized methods would be expected to achieve higher test accuracy. As for generic predictors, some methods are trained directly from data, while others, such as CanDrA and TransFIC, use predictions from pre-existing variant effect predictors

Name	Method and features used	Reference
CHASM	Random Forest method using evolutionary and structural features.	Carter et al. [29] Tokheim et al. [41]
CRAVAT4	An evolving suite of informatics tools for mutation interpretation and impact prediction.	Masica et al. [38]
CScape	Gradient boosting (sequential learner) using evolutionary and genomic features	Rogers et al. [42]
CScape-somatic	Similar to CScape except distinguishes rare from recurrent somatic SNVs using cancer data only.	Rogers et al. [43]
FATHMM-cancer	Using evolutionary data, a predecessor to CScape.	Shihab et al. [44]
FunSeq2	Scoring scheme, using conservation, regulatory and other measures. Prioritizes cancer somatic variants, especially for regulatory noncoding mutations.	Fu et al. [45]
CanDrA	Support Vector Machine method using 10 published predictors (CHASM, SIFT and others) and evolutionary, structural and gene features.	Mao et al. [73]
TransFIC	Scoring method utlilizing SIFT, Polyphen2 and MutationAssessor	Gonzalez-Perez [47]