Table 1

Results of variant classification for individual tool, and two consensus-based combinations, for the (A) open (n=8480) and (B) clinical (n=1757) datasets

True positiveTrue negativeFalse positiveFalse negativeSensitivitySpecificityMCCLR+LR−
(A) Open dataset
IndividualSIFT2302385718784430.840.670.482.6:11:4.2
PolyPhen-22387417715583580.870.730.563.2:11:5.6
REVEL239454452903510.870.950.8317.2:11:7.4
GAVIN261556111241300.950.980.9344.1:11:20.7
ClinPred2469573142760.901.000.931289.6:11:9.9
ConsensusSIFT+PolyPhen-22240341023255050.820.590.392:11:3.2
REVEL+ClinPred223354422935120.810.950.7815.9:11:5.1
(B) Clinical dataset
IndividualSIFT10312124061080.910.340.311.38:11:3.62
PolyPhen-210212114071180.900.340.291.36:11:3.3
REVEL9833702481560.860.600.482.15:11:4.37
GAVIN1100157461390.970.250.331.29:11:7.42
ClinPred1107167451320.970.270.361.33:11:9.62
ConsensusSIFT+PolyPhen-29601354831790.840.220.081.08:11:1.39
REVEL+ClinPred9731424761660.850.230.111.11:11:1.58
  • For consensus-based results, non-concordant, where tools disagree on the classification, were considered incorrect. Matthews correlation coefficient (MCC) was calculated as follows:

  • LR+ is the positive likelihood ratio; LR− is the negative likelihood ratio.

  • FN, false negatives (ie, pathogenic variants predicted to be benign); FP, false positives (ie, benign variants predicted to be pathogenic); TN, true negatives (ie, benign variants predicted to be benign); TP, true positives (ie, pathogenic variants predicted to be pathogenic).