DegreeCox – a network-based regularization method for survival analysis

BMC Bioinformatics

Table 1 Deviance and C-index 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 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

	Train	Bonome						TCGA						Tothill
	Test	Bonome		TCGA		Tothill		Bonome		TCGA		Tothill		Bonome		TCGA		Tothill
	Network	GCN	GFM	GCN	GFM	GCN	GFM	GCN	GFM	GCN	GFM	GCN	GFM	GCN	GFM	GCN	GFM	GCN	GFM
RMSE	DegreeCox	0 . 5 5 8 1	0 . 7 7 2 4	1.3189	1.1538	1.2139	1.1027	1 . 2 3 6 7	1.3619	0.9201	0.8043	1 . 0 5 7 3	1.1083	1.6326	1.2975	1.3749	1.1679	0 . 5 1 1 6	0.7013
	Net-Cox	0.8131	0.8353	1.1438	1 . 1 2 8 5	1.0992	1 . 0 8 8 6	1.3514	1 . 3 0 4 5	0.8361	0.8508	1.1003	1 . 0 8 0 2	1 . 2 9 1 7	1 . 2 5 9 1	1 . 1 6 1 2	1 . 1 4 0 3	0.7363	0.7606
	Ridge	0.7807		1 . 1 4 1 3		1 . 0 9 8 6		1.3755		0 . 7 2 1 5		1.1769		1.5649		1.3252			0 . 5 4 3 2
	Lasso	0.7887		1.4619		1.2586		1.7419		0.8105		1.3019		1.9595		1.4208		0.5444
C-Index	DegreeCox	0 . 9 7 9 5	0.9401	0.6020	0.6037	0.6455	0.6494	0.6444	0.6427	0.8476	0.9089	0 . 6 7 1 1	0.6695	0.6011	0.6088	0.6100	0.6215	0 . 9 8 3 4	0.9519
	Net-Cox	0.9260	0.9202	0.6079	0.6054	0.6483	0.6506	0.6416	0.6439	0.8918	0.8892	0.6633	0 . 6 7 0 5	0 . 6 1 5 2	0 . 6 1 0 6	0 . 6 2 4 4	0 . 6 2 5 0	0.9389	0.9352
	Ridge		0 . 9 4 1 0	0 . 6 1 7 7		0 . 6 5 6 9		0 . 6 4 9 2		0 . 9 3 9 4		0.6579		0.6000		0.5926			0 . 9 8 2 9
	Lasso	0.9309		0.5615		0.6124		0.6405		0.9043		0.6399		0.5075		0.5728		0.9784

Values in bold represent the best performing method for the dataset/network combination (per RMSE and C-Index)

ISSN: 1471-2105