Table 4.
P-value of the log-rank test of lower dimensional representation, generated by Partitioning Around Medoids (PAM) clustering algorithm on the last hidden layers of three Deep Learning-based approaches (testing set only). 12 TCGA (The Cancer Genome Atlas) cancers are being compared. Bolded values indicate the smallest p-value among three Deep Learning approaches, refer to better low dimensional representation
| TCGA Cancers | P-values of log-rank test by models | ||
|---|---|---|---|
| Cox-nnet | DeepSurv | AECOX | |
| KIRP | 7.71E-02 | 2.45E-01 | 1.40E-01 |
| KIRC | 2.38E-03 | 9.79E-02 | 3.01E-01 |
| LIHC | 3.57E-01 | 6.14E-01 | 3.86E-01 |
| BRCA | 4.45E-01 | 4.81E-01 | 4.85E-01 |
| CESC | 2.45E-01 | 3.26E-01 | 3.92E-01 |
| LUAD | 9.58E-02 | 4.07E-01 | 1.11E-01 |
| BLCA | 3.27E-01 | 2.67E-01 | 4.94E-01 |
| HNSC | 3.38E-01 | 4.06E-01 | 6.19E-01 |
| PAAD | 2.97E-01 | 4.04E-01 | 2.29E-01 |
| OV | 2.80E-01 | 3.38E-01 | 4.42E-01 |
| STAD | 5.72E-01 | 2.67E-01 | 7.01E-01 |
| LUSC | 3.10E-01 | 4.14E-01 | 6.05E-01 |
The boldface p-value indicates it is the smallest one among all three algorithms