Table 3. USC training results.

MOLECULAR CLASSIFIERS
	MODEL 1	MODEL 2
#Mistakes	0	0
Delta	1	0.5
Rho	0.7	0.7
Average Genes	18	34
Predictive Genes	ASS1, BAP1, CAV1, CCNB1, CD44, CDH1, EGR3, FN1, ITGA3, KRT5, LAMA3, LGALS3, MICAL2, MMP9, MYH11, NME2, NMU, PAPPA, PECAM1, PKM, RAD21, TGFBR2	ASS1, BAP1, CAV1, CCNB1, CD44, CDH1, CDH11, COL4A2, CTNNA1, CXADR, EEF2, EGR3, EIF4G1, FANCI, FN1, GALNT7, GLI2, HEG1, IFITM1, ITGA3, KRT5, LAMA3, LGALS3, MAGED1, MICAL2, MMP9, MYH11, NME2, NMU, PAK4, PAPPA, PECAM1, PKM, PTGS2, RAD21, SDC1, SMARCA4, TGFBR2, TOP2A, VEGFA

USC classifiers and parameters: #Mistakes: number of classification mistakes in the training phase; Delta: shrunken threshold; Rho: correlation threshold; Average genes: average number of genes, which are selected among the predictive ones, used for classification; Predictive genes: genes included in each classifier, selected among the 117 gave as input