TABLE 5

Details on the RRCK data modeling grid, i.e., performance of various modeling methods versus descriptors

Some combinations were not evaluated because it was apparent from the combination of CDK descriptors and the modeling methods that the results were not going to be equivalent to or better than the baseline model.

SVMRP Forest Uni ClassRP ForestC5.0
CDK
    κ0.120.180.150.47
    Sensitivity0.110.440.540.59
    Specificity0.980.770.650.93
    PPV0.510.300.260.67
MOE2D and SMARTS KeysNot evaluatedNot evaluatedNot evaluated
    κ0.53
    Sensitivity0.64
    Specificity0.94
    PPV0.72 (baseline)
CDK and SMARTS KeysNot evaluatedNot evaluatedNot evaluated
    κ0.50
    Sensitivity0.62
    Specificity0.94
    PPV0.68
  • PPV, positive predicted value.