Table 2. This table contains the results of our feature selection procedure for miRs profiled in plasma.

Rank	Top 1	Top 2	Top 3	Top 4	Top 5	Top 6	Top 7	Top 8	miR
T-test
1	33	37	47	53	60	67	70	70	miR16
2	13	37	47	53	57	57	57	63	miR145
3	17	40	40	50	53	60	67	73	miR106b
4	27	37	40	43	50	50	57	63	miR34a
5	3	10	27	33	37	43	57	70	miR182
6	0	0	13	30	43	50	57	63	miR20a
7	3	7	23	23	43	47	53	57	miR21
8	0	10	23	30	43	57	70	83	miR222
9	0	10	10	27	27	40	53	73	miR331
10	0	7	10	13	20	40	47	57	miR375
11	3	3	13	27	37	47	60	63	miR205
12	0	3	7	17	30	43	53	63	miR221
Centroid
1	17	27	37	47	63	70	80	80	miR16
2	23	30	37	50	50	57	57	67	miR106b
3	13	33	57	63	70	70	73	73	miR145
4	13	30	33	37	40	47	60	60	miR34a
5	0	0	3	13	20	23	47	70	miR20a
6	7	10	10	20	23	40	43	60	miR182
7	0	7	30	37	40	50	63	80	miR21
8	0	3	10	23	37	47	47	57	miR222
9	13	33	40	53	60	63	73	73	miR331
10	3	7	10	13	27	43	53	63	miR375
11	10	20	30	37	53	63	70	77	miR221
12	0	0	3	7	17	27	33	40	miR205

The miRs have been ranked based on their appearance in the Top n miRs for different classifiers as part of the cross-validation procedure. The percentage of times the miR is selected by the classifier in the top n = 1..8 miRs is reported for both t-test and centroid approaches.