Skip to main content

Table 3 The ROC score of methods for multi-modal data. NN is neural network model, NN 1/0 is neural network model that use binary representation of GO terms as features. (concat) is approach where GO terms and k-mer is concatenated as single vector to represent each protein, (multi-input) approach where GO terms and k-mer are used as two input to the model. The number of layers in neural network are three and the dimension of neurons in each layer are 128,64, and 1

From: Predicting metabolic pathway membership with deep neural networks by integrating sequential and ontology information

Methods BP+k-mer CC+k-mer MF+k-mer
NN (concat) 0.954 0.874 0.907
NN (multi-input) 0.957 0.880 0.907
NN 1/0 (concat) 0.943 0.883 0.894
NN 1/0 (multi-input) 0.940 0.882 0.888
cosine (concat) 0.920 0.768 0.687
SVM (concat) 0.933 0.814 0.863
RF (concat) 0.923 0.840 0.844
KNN (concat) 0.829 0.784 0.790
  1. not significantly different than cosine method in each ontology