Table 6. SVM Misclassification Tables.

A: WB trained SVM applied to PBMC accurately predicts MED and HIGH, with confusion of CTRL and LOW.
	CTRL	LOW	MED	HIGH
CTRL	1	6	0	0
LOW	2	5	1	0
MED	0	0	14	0
HIGH	0	0	0	8
B: PBMC trained SVM applied to WB is not good as expected, because the SVM was trained on PBMC genes that were not found to be significant in WB.
CTRL	2	0	0	1
LOW	0	8	0	0
MED	3	0	10	1
HIGH	4	0	1	3
C: Overlapping PBMC SVM trained applied to PBMC (This means that the subset of genes in PBMC that are the overlapping genes with WB are sufficient in reproducing the prediction found by the entire set)
CTRL	7	0	0	0
LOW	0	8	0	0
MED	0	0	14	0
HIGH	0	0	0	8
D: Overlapping PBMC SVM trained applied to WB. (This shows that the prediction of the MED and HIGH risk groups based on training the SVM on the Overlap PBMC gene set is transferrable to the WB dataset. This is the consistency argument we are looking for.)
CTRL	3	0	0	0
LOW	3	3	0	2
MED	0	0	14	0
HIGH	0	0	1	7