Skip to main content

Table 4 AUC calculated for classification sets (higher values are better)

From: Transformer-CNN: Swiss knife for QSAR modeling and interpretation

DatasetDescriptor based methodsaSMILES based (augm = 10)2Transformer-CNN, no augmTransformer-CNN, augm = 10CDDD descriptorsb
HIV0.820.780.810.830.74
AMES0.860.880.860.890.86
BACE0.880.890.890.910.9
Clintox0.77 ± 0.030.76 ± 0.030.71 ± 0.020.77 ± 0.020.73 ± 0.02
Tox210.790.830.810.820.82
BBBP0.900.910.90.920.89
JAK30.79 ± 0.020.8 ± 0.020.70 ± 0.020.78 ± 0.020.76 ± 0.02
BioDeg0.920.930.910.930.92
RP AR0.850.870.830.870.86
  1. We omitted the standard mean errors, which are 0.01 or less, for the reported values
  2. aResults from our previous study [22]. bBest performance calculated with CDDD descriptors obtained using Sml2canSml autoencoder from [27]