Skip to main content
. 2022 Jan 28;11(2):209. doi: 10.3390/biology11020209

Table 5.

Performance of the RF and DNN QSAR classification models on the external test sets. Consensus models built using the average probability of each active chemical.

PubChem (ID: AID588834) (135 Actives and 876 Inactives)
TOX21 Model Q Qb Sp Se MCC
DNN 0.665 0.676 0.661 0.690 0.245
RF 0.920 0.793 0.966 0.619 0.631
Consensus 0.873 0.788 0.904 0.673 0.518
TOX21-ChEMBL
DNN 0.693 0.703 0.690 0.717 0.286
RF 0.926 0.766 0.984 0.549 0.641
Combined (RF + DNN) 0.900 0.791 0.945 0.637 0.582
Lit-based hERG inhibitors (393 actives)
TOX21 model Q TP FN
DNN 0.41 161 231
RF 0.40 157 235
Combined (RF + DNN) 0.40 157 235
TOX21-ChEMBL
DNN 0.389 153 249
RF 0.341 134 258
Combined (RF + DNN) 0.341 134 258