Skip to main content

Table 4 Model performance measures for ChEMBL1614027 based on DTC-split model 1 and all-split model 5

From: Nonadditivity in public and inhouse data: implications for drug design

 

RF

SVM

PLS

Train R2 (RMSE)

Test R2 (RMSE)

Test MCC

Train R2 (RMSE)

Test R2 (RMSE)

Test MCC

Train R2 (RMSE)

Test R2 (RMSE)

A*

NA#

A*

NA#

A*

NA#

A*

NA#

A*

NA#

DTC-split

0.93 (0.15)

0.60 (0.36)

− 0.48 (1.26)

0.62

0.28

0.84 (0.23)

0.60 (0.36)

− 0.46 (1.25)

0.69

0.31

0.76 (0.28)

0.54 (0.39)

− 0.60 (1.31)

All-split

0.78 (0.33)

0.34 (0.57)

− 0.35 (1.20)

0.40

0.22

0.49 (0.50)

0.32 (0.57)

− 0.43 (1.23)

0.47

0.32

0.45 (0.52)

0.25 (0.60)

− 0.39 (1.22)

  1. Bold values are best performance measures across DTC-split and All-split and across different ML approaches
  2. Train R2 is based on 5-fold cross validation results
  3. *Additive test data
  4. #Nonadditive test data.