Skip to main content
. 2020 Feb 19;60(3):1215–1223. doi: 10.1021/acs.jcim.9b01057

Table 3. Classification Performance of the Consensus Approaches Estimated on the Binding, Antagonism, and Agonism Sets Considering the Best Five Models Only (Selected Based on NER)a.

  binding (5 models)
antagonism (5 models)
agonism (5 models)
consensus approach Sn (%) Sp (%) NER (%) Cvg (%) rank Sn (%) Sp (%) NER (%) Cvg (%) rank Sn (%) Sp (%) NER (%) Cvg (%) rank
MVL 63.9 97.7 80.8 92.8 3 71.6 82.8 77.2 96.5 2 73.8 98.8 86.3 98.6 2
MVI 65.7 99.3 82.5 88.4 4 71.9 84.4 78.1 90.9 3 74.1 99.0 86.5 97.0 4
MVS 63.8 99.5 81.6 87.4 6 78.3 87.6 83.0 74.4 5 73.1 99.2 86.1 95.8 6
B 72.0 91.0 81.5 100 1 73.2 79.7 76.5 100 1 74.4 96.1 85.2 99.9 3
Bp 88.3 96.2 92.2 58.6 7 79.4 85.5 82.4 76.3 4 76.1 98.2 87.1 90.0 7
a

For each consensus approach, sensitivity (Sn), specificity (Sp), non-error rate (NER), coverage (Cvg), and total ranking are reported. B, Bayes; Bp, protective Bayes; MVL, majority voting loose; MVI, majority voting intermediate; MVS, majority voting strict.