Skip to main content

Table 10 Ablation analysis of base learner interactions in stacking meta classifiers

From: A meta-learning approach for B-cell conformational epitope prediction

Classifier*

TPR

FPR

Precision

Accuracy

F-score

MCC

AUC

3-level stacking

0.194

0.008

0.520

0.958

0.283

0.300

0.793

\SEPPA 2.0

0.169

0.009

0.447

0.956

0.245

0.256

0.755

\ElliPro

0.144

0.012

0.349

0.952

0.204

0.203

0.746

\BepiPred

0.144

0.012

0.341

0.952

0.203

0.200

0.749

\BCPREDS

0.109

0.009

0.338

0.953

0.165

0.173

0.717

\AAP

0.124

0.016

0.255

0.947

0.167

0.153

0.758

\Bpredictor

0.154

0.012

0.356

0.952

0.215

0.213

0.724

\DiscoTope 2.0

0.045

0.006

0.257

0.954

0.076

0.092

0.672

\ABCpred

0.065

0.006

0.342

0.955

0.109

0.134

0.658

  1. *Classifiers tested in the ablation analysis. The first classifier in the first row is the stacking meta classifier that employs all of the 8 base learners (Figure 3). The remaining classifiers are listed in the order in which they were selected to be removed iteratively from the stacking meta classifier for the ablation study. `\’ indicates removed. For example, the second classifier is the stacking meta classifier after SEPPA 2.0 was removed, and the third classifier is the stacking meta classifier after SEPPA 2.0 and ElliPro were removed from the meta model. The meta classifier in the final row did not apply any base learner after the final prediction tool ABCpred was removed.