Skip to main content

Table 1 Performance summary of CEEdge and PPIEdge feature types compared to gene-based feature type in RF models. “Win” indicates that the edge-based feature type outperformed gene-based feature type and “Lose” indicates that the gene-based feature type outperformed edge-based one. “SS” and “NSS” indicate that win/lose was statistically significant and not statistically significant respectively

From: Robust edge-based biomarker discovery improves prediction of breast cancer metastasis

Evaluation Metric Type Feature Win Lose Win Lose Equal
  Type SS SS NSS NSS /NaN
AUC CEEDge 4 0 7 2 0
  PPIEdge 7 4 1 1 0
F1-score (0.5 probability threshold) CEEdge 12 0 0 0 1
  PPIEdge 10 0 1 0 2
Kappa (0.5 probability threshold) CEEdge 11 0 1 0 1
  PPIEdge 8 0 2 1 2
F1-score (Optimal probability threshold) CEEdge 4 1 6 2 0
  PPIEdge 7 1 1 4 0
Kappa (Optimal probability threshold) CEEdge 4 0 6 2 1
  PPIEdge 6 0 3 3 1