Skip to main content

Table 9 Results of different network architectures on S. cerevisiae, the adopted negative dataset is constructed by NIP-SS

From: Predicting protein-protein interactions using high-quality non-interacting pairs

Architectures Data set ACC PE RE SPE MCC F 1 AUC
DNNs Fold 1 94.08% 94.04% 94.17% 93.98% 88.15% 94.11% 98.24%
  Fold 2 94.03% 94.36% 93.64% 94.42% 88.07% 94.00% 98.13%
  Fold 3 94.57% 95.25% 93.66% 95.45% 89.14% 94.45% 98.17%
  Fold 4 94.38% 94.99% 93.78% 94.98% 88.77% 94.38% 98.16%
  Fold 5 94.19% 94.84% 93.50% 94.88% 88.39% 94.17% 98.03%
  Average 94.25% ± 0.22% 94.70% ± 0.49% 93.75% ± 0.26% 94.74% ± 0.56% 88.5% ± 0.45% 94.22% ± 0.19% 98.15% ± 0.08%
DNNs-Con Fold 1 91.92% 92.40% 91.27% 92.55% 83.84% 91.83% 97.15%
  Fold 2 91.86% 93.87% 89.21% 94.40% 83.79% 91.48% 96.90%
  Fold 3 91.58% 93.62% 89.32% 93.86% 83.26% 91.42% 96.83%
  Fold 4 91.86% 93.65% 90.07% 93.70% 83.79% 91.83% 96.92%
  Fold 5 91.42% 92.24% 90.53% 92.32% 82.86% 91.38% 96.93%
  Average 91.73% ± 0.21% ∙ 93.16% ± 0.77% ∙ 90.08% ± 0.86% ∙ 93.37% ± 0.89% ∙ 83.51% ± 0.43% ∙ 91.59% ± 0.23% ∙ 96.95% ± 0.12% ∙
  1. ∙/ indicates whether our model is statistically (according to pairwise t-test at 95% significance level) superior/inferior to the DNNs-Con