Table 1.

Fold difference in the expression values of informative probe sets for 3 different clinical prediction problems assessed in the same breast cancer data set (GEO GSE 16716)

	ER+ versus ER-1		pCR versus RD2		pCR versus RD in ER- cancers only3
	Feature # 10	Feature # 100	Feature # 10	Feature # 100	Feature # 10	Feature # 100
FDR adjusted p-value	4.71E-12	3.26E-07	0.004	0.0205	0.4	0.68
Fold difference4
< 0.5	0	0	0	6	4	59
≥0.5 - < 1.0	0	13	3	43	2	32
≥1.0 - < 1.2	0	8	1	15	4	7
≥1.2 - < 1.5	0	18	2	17	0	1
≥1.5 - < 2.0	0	18	2	11	0	1
≥ 2.0 - < 3.0	6	28	2	8	0	0
≥ 3.0 - < 4.0	3	12	0	0	0	0
≥ 4.0	1	3	0	0	0	0

¹Random sample of 41 ER Positive and 50 ER Negative Samples

²Random sample of 41 pathologic CR and 50 residual cancers regardless of ER status.

³Random sample of 41 pathologic CR and 50 residual cancers all ER Negative.

⁴Log₂ Difference = abs(mean log₂ intensity for group1 - mean log₂ intensity for group 2) where "abs" is absolute value.

The numbers of probe sets with a given level of differential expression are shown for the 3 comparisons including (i) Estrogen Receptor (ER)-positive versus ER-negative cancers, (ii) cancers with pathologic complete response (pCR) to chemotherapy versus lesser response (RD) and (iii) cases with pCR versus RD in ER-negative cancers only. Probe sets with mean log₂ transformed expression difference >1 between comparisons groups are highlighted in bold. FDR = false discovery rate (i.e., proportion of genes detected to be informative which are not truly informative). FDR adjusted p-values (also known as FDR q-values) are the estimated FDR values that would be incurred if the p-values associated with the selected genes were used as the threshold for significance (i.e., genes with that p-value or a lower p-value were to be detected as informative). So for the Feature #10 column, the reported FDR adjusted p-value is the q-value associated with the 10th highest ranked gene.