Skip to main content

Table 5 Performance by various propensity integration methods

From: Staged heterogeneity learning to identify conformational B-cell epitopes from antigen sequences

Group

Methods

Propensities

Recall

Precision

Fscore

Voting

Ranked votinga

209

0.650

0.191

0.295

 

Exhaustive voting

5 groups

0.656

0.191

0.295

General

SVM

5 groups

0.614

0.181

0.279

 

Random forest

5 groups

0.600

0.186

0.284

Bayesian

Naïve Bayesian

5 groups

0.686

0.193

0.301

 

Bayesian network

5 groups

0.775

0.173

0.283

Tree

Decision tree

5 groups

0.675

0.196

0.304

 

Regression tree

5 groups

0.655

0.194

0.299

General

SVM

209

0.594

0.191

0.289

 

Random forest

209

0.639

0.191

0.294

Bayesian

Naïve Bayesian

209

0.633

0.198

0.301

 

Bayesian network

209

0.586

0.170

0.264

Tree

Decision tree

209

0.633

0.198

0.301

  1. The parameters used in these models are tuned to realize an optimal performance. The best ranked voting was achieved by propensity PSSM (P209), ASA (P206) and optimized beta-structure-coil equilibrium constant (P170, AAIndex ID: OOBM850101). The best exhaustive voting results was realized by a combination of PSSM, physico-chemical propensities and ASA
  2. aDue to the complexity of exhaustive voting algorithm, only five groups of propensities were used. Regression tree constructed on 209 propensities required too much computation time, and no results were obtained