Skip to main content

Table 2 Comparison of DDIs prediction performance on LSTM and DNN model. The p value compared with LSTM is added in brackets

From: Novel deep learning-based transcriptome data analysis for drug-drug interaction prediction with an application in diabetes

Feature

Method

Macro-F1

Macro-recall

Macro-precision

Original

DNN

90% ± 1.9% (0.0008)

90.7% ± 1.8% (0.0007)

91.3% ± 2.3% (0.0056)

 

LSTM

94.2% ± 1.9% (–)

95.5% ± 1.9% (–)

93.5% ± 1.9% (–)

Autoencoder

DNN

91.2% ± 0.7% (0.086)

90.8% ± 0.9% (0.0013)

93.2% ± 1.1% (0.0445)

 

LSTM

92.5% ± 1.5% (–)

95.2% ± 1.6% (–)

90.8% ± 1.6% (–)

GCAN

DNN

93.3% ± 1.4% (0.004)

93.9% ± 1.7% (0.008)

93.7% ± 1.4% (0.12)

 

LSTM

95.3% ± 1.5% (–)

96.6% ± 1.3% (–)

94.6% ± 1.9% (–)

  1. Bold indicates the best prediction performance