Table 1 DeMAG outperforms other VEPs on clinical variants

From: DeMAG predicts the effects of variants in clinically actionable genes by integrating structural and evolutionary epistatic features

VEPs

Sensitivity

Specificity

Accuracy

MCC

AUC

Variants predicted (n = 1285)

VEP’s coverage

DeMAG

0.93

0.82

0.90

0.77

0.96

–

–

SIFT4G

1

0.07

0.68

0.21

0.81

1226

0.954

DeMAG

0.94

0.82

0.89

0.76

0.96

–

–

REVEL

0.99

0.62

0.86

0.69

0.96

1285

1

DeMAG

0.93

0.79

0.89

0.74

0.95

–

–

DEOGEN2

0.91

0.58

0.80

0.53

0.91

964

0.750

DeMAG

0.93

0.79

0.89

0.74

0.95

–

–

PolyPhen2

0.91

0.72

0.85

0.64

0.86

972

0.756

DeMAG

0.94

0.81

0.89

0.76

0.96

–

–

VEST4

0.99

0.64

0.87

0.72

0.95

1280

0.996

DeMAG

0.94

0.81

0.90

0.76

0.96

–

–

M-CAP

1

0.28

0.77

0.44

0.93

1256

0.977

DeMAGa/b

0.96/0.96

0.81/0.80

0.92/0.91

0.80/0.78

0.97/0.96

–

–

EVEa/b

0.92/0.86

0.81/0.73

0.89/0.82

0.73/0.58

0.91/0.89

940/1165

0.731/0.906

  1. Different performance metrics for DeMAG and seven popular variant effect predictors (VEPs). The test set is assembled from the ClinVar database, consisting of both pathogenic (n = 852) and benign variants (n = 433) submitted after the year 2017 (see Methods). The comparison in pairs guarantees that each predictor is evaluated on all the variants for which a prediction exists. DeMAG has 100% coverage and it is the most balanced across all the metrics. The comparison with EVE includes two values. The first value indicates the performance on the variants which are predicted by EVE as benign and pathogenic excluding the uncertain class (73% of all variants). The second value includes variants that EVE misclassifies as uncertain. We assigned them as if EVE was a random classifier.
  2. aIf EVE’s Uncertain class variants are excluded from the testing set.
  3. bIf EVE’s Uncertain class variants are assigned as if EVE was a random classifier.
Back to article page