Skip to main content
. 2018 Sep 19;3(9):11392–11406. doi: 10.1021/acsomega.8b01647

Table 5. Results for PBT Index of Chemicals (Model Dataset 5) with Best Weighting (0.5–0–0.5) Combination as Obtained from the Retrospective Study.

model division method R2 Q2 MAE95%Train QExt-F12 MAE95%Test correct prediction in % training set range number of compounds outside ADa
1 sorted responseb 0.88 0.87 0.36 0.92 0.37 91.67 8.1 1 (189)
2   0.90 0.89 0.33 0.95 0.27 100   2 (41, 189)
3   0.89 0.88 0.35 0.94 0.30 100   2 (14, 189)
4   0.89 0.88 0.34 0.93 0.34 91.67   1 (14)
5   0.89 0.88 0.34 0.94 0.32 88.89   4 (14, 41, 189, 206)
6 Kennard–Stonec 0.91 0.90 0.35 0.87 0.34 95.56 8.1 2 (211, 212)
7   0.90 0.89 0.34 0.88 0.29 95.56   0
8   0.91 0.90 0.34 0.88 0.34 95.56   0
9   0.91 0.90 0.34 0.88 0.31 95.56   0
10   0.92 0.91 0.33 0.88 0.31 95.56   0
11 modified-k-medoidsd 0.91 0.90 0.32 0.84 0.45 88.64 7.24 3 (11, 25, 189)
12   0.90 0.89 0.32 0.87 0.40 86.36   2 (11, 25)
13   0.92 0.91 0.31 0.86 0.43 93.18   2 (11, 25)
14   0.90 0.89 0.32 0.88 0.39 93.18   0
15   0.92 0.91 0.31 0.86 0.43 93.18   2 (11, 25)
a

AD using standardization technique is used and compound ID mentioned under parenthesis.

b

NTraining = 144, NTest = 36.

c

NTraining = 135, NTest = 45.

d

NTraining = 136, NTest = 44.