Fig. 9.

Prediction performance comparison of our proposed method with other predictors. The MTT-based prediction model was compared with two recently proposed compound-target interaction models in our released natural product compound-target interaction dataset. DeepPurpose(CNN, CNN) represents the prediction model in [24] with a Convolutional neural network as feature encoders both for compounds and proteins. Other DeepPurpose symbols were interpreted similarly. Ten-fold cross-validation was used to evaluate the prediction performance of the various compared C–T models