Skip to main content

Table 7 Comparison of our GCN models (SL-GCN and SSL-GCN) and the models constructed using the DeepChem built-in ML methods

From: Chemical toxicity prediction based on semi-supervised learning and graph convolutional neural network

Model Description Overall score Std. Refs.
logreg Logistic regression model 0.6397 [52]
tf Deep neural network 0.6582 0.0097 [37]
tf-robust Deep neural network (with bypass layers) 0.6825 0.0056 [53]
rf Random forest model 0.6618 0.0066 [52]
kernelsvm Kernel SVM model 0.7000 [52]
graphconv Graph convolutional model 0.6943 0.0043 [54]
irv Influence relevance voting (IRV) classifier 0.6853 [55]
xgb Xgboost classification model 0.6908 0.0039 [56]
SL-GCN Supervised GCN model 0.7156 0.0068 This study
SSL-GCN Semi-supervised GCN model 0.7571 0.0084 This study
  1. The overall score is the average ROC-AUC score in predicting the 12 prediction tasks in the test set. The experiments were repeated 5 times
  2. The bold number denotes the best overall score among all models