Skip to main content

Table 2 P-values for log-rank test results for models chosen by 5-fold cross-validation and tested on all 3 datasets (including 2 that were hidden from the training phase). The log-rank tests the separation in two categories of patients, high and low risk based on the expression dataset, using the top and lower 40 % PI groups and the top and lower 50 % PI groups. The LASSO and RIDGE methods do not use network information so the values for GCN and GFM are the same, they are only shown in both networks when they are better than DEGREECOX and NET-COX. The p-values when the model is tested on the same dataset used in training are always 0 and are ommited from the table

From: DegreeCox – a network-based regularization method for survival analysis

  Train Bonome TCGA Tothill
  Test TCGA Tothill Bonome Tothill Bonome TCGA
  Network GCN GFM GCN GFM GCN GFM GCN GFM GCN GFM GCN GFM
40 %PI Thres. DegreeCox 2.084E-5 2.124E-5 0.0013 4.390E-4 2.046E-4 3.990E-4 6 . 5 4 7 E 6 5.822E-6 9 . 8 3 3 E-5 1 . 7 5 7 E-4 8.347E-8 3.125E-8
  Net-Cox 1.082E-4 2.791E-5 7.726E-4 1 . 5 1 4 E-4 2.815E-4 1.185E-4 4.241E-5 1 . 4 3 2 E-6 1.696E-4 2.545E-4 7 . 7 1 7 E-9 4 . 5 0 3 E-9
  Ridge 1 . 5 9 4 E-6 2 . 5 3 7 E-4   4.233E-4 1.765E-5 0.0016 1.864E-5
  Lasso 0.0364 0.0048 7 . 4 3 6 E-5 0.0036 0.5630 0.0033
50 %PI Thres. DegreeCox 3.332E-4 5.284E-5 0.0076 0.0084 4.394E-4 0.0090 5 . 7 8 1 E-5 1 . 3 0 9 E-4 0.0045 4 . 3 0 2 E-4 5.264E-7 7.183E-7
  Net-Cox 2.169E-5 5.086E-5 0.0170 0.0179 0.0036 0.0015 1.247E-4 3.126E-4 0 . 0 0 2 6 8.138E-4 1 . 1 0 5 E-8 1 . 6 3 2 E-7
  Ridge 1 . 7 9 5 E-5 0 . 0 0 1 3 3 . 1 9 3 E-4 0.0029 0.0050 3.499E-5
  Lasso 0.0720 0.0048 0.0022 0.0193 0.6464 0.0050
  1. Values in bold represent the best method for the dataset/network combination (per 40 % and 50 % separation)