Skip to main content

Table 9 Comparison of the predictive performance of TL and STP

From: Predicting anatomic therapeutic chemical classification codes using tiered learning

Comparison of CA(k) using STP and TL at the second, third, and fourth ATC levels

 

STP N(2)

N(2)

SuperPred

(+/−)

N(2)

ChEMBL

(+/−)

STP CA(3)

N(3)

SuperPred

(+/−)

N(3)

ChEMBL

(+/−)

STP CA(4)

N(4)

SuperPred

(+/−)

N(4)

ChEMBL

(+/−)

MLP

377

382

(+5)

383

(+6)

330

325

(−5)

372

(+42)

225

207

(−18)

234

(+9)

Naive Bayes

203

320

(+117)

363

(+160)

170

260

(+90)

293

(+123)

127

162

(+35)

183

(+56)

Random Forest

329

371

(+42)

431

(+102)

280

306

(+26)

348

(+68)

159

189

(+30)

215

(+56)

SVM

360

395

(+35)

457

(+97)

327

354

(+27)

400

(+73)

230

253

(+23)

277

(+47)

  1. This table compares the number of compounds underlying N(k) for each classifier in predicting the second through fourth level of the ATC code using TL and STP. Only the TL initializations using SuperPred and ChEMBL and shown. The change (+/−) is noted below the numbers underlying N(k) for each level. The best results are highlighted for each initialization strategy