Stage II and stage III colorectal cancer

Total of 111 microarray datasets (77 for LCM samples, and 17 pairs for homogenized samples from tumor and adjacent tissues) were normalized using robust multi-array average (RMA) method under R 2.6.2 statistical software together with BioConductor package, as described previously. Then, the gene expression levels were log2-transformed, and 62 control probe sets were removed for further analysis. In order to identify a set of genes associated with development of metastatic recurrence, we performed Wilcoxon rank-sum test for gene expression differences of 54,613 probe sets between recurrence and non-recurrence groups. Similarly, Wilcoxon singed-rank test was conducted to select genes which showed significant expression difference between tumor and adjacent tissue. Then, we selected a set of genes that satisfied both of above two criteria.