Skip to main content

Table 4 Performance of the hERG prediction models derived from different combinations of ML algorithms and descriptor sets

From: Semi-automated workflow for molecular pair analysis and QSAR-assisted transformation space expansion

 

Fivefold cross-validation

Test set

SE

SP

ACC

AUC

SE

SP

ACC

AUC

MACCS

 GB

0.873

0.84

0.857

0.925

0.883

0.860

0.871

0.939

 RF

0.857

0.803

0.830

0.902

0.871

0.809

0.840

0.915

 SVM

0.836

0.851

0.843

0.914

0.846

0.861

0.854

0.929

 XGBoost

0.874

0.84

0.857

0.926

0.880

0.863

0.872

0.939

MOE2D

 GB

0.878

0.855

0.866

0.934

0.896

0.865

0.880

0.942

 RF

0.872

0.844

0.858

0.925

0.883

0.839

0.861

0.935

 SVM

0.883

0.856

0.869

0.936

0.854

0.841

0.848

0.920

 XGBoost

0.883

0.856

0.869

0.936

0.887

0.866

0.877

0.945

Consensus model

0.865

0.882

0.874

0.935

0.894

0.865

0.879

0.946