From a total of 1475 unique, annotated genes identified in 23 independent GEP studies, 124 genes were reported in at least two studies, and only 9 of them in three studies, which give us a clear idea of the lack of reproducibility at the individual gene level. This lack of reproducibility does not seem to be caused by the different investigated features related to cancer prognosis,SB-258719 since the proportion of genes reported by two studies of the same class was even lower than for all studies together. Unexpectedly, 70 out of these 124 genes showed contrasting direction in expression change between two single studies, while for the other 54 the expression change was in the same direction, 19 up-regulated and 35 down-regulated. The proportion of upand down regulated genes was approximately the same also within each of the consistently enriched GO and KEGG categories. The inconsistencies in the direction of differential expression can be attributed to several factors: first, the large number of false positives observed in microarray gene expression studies ; second, overgeneralization of comparisons in metaanalyses,BIQ especially related to experimental design and background reference for expression; third, heterogeneity in the tissue samples used in each study; and fourth, inaccurate results due to poor study design. However, a clear explanation for these discrepancies is missing. Only one previous meta-analysis of ten GEP studies has reported a list of 13 genes differentially expressed in CRC with good versus bad prognosis, reported by at least two independent studies. A comparison with our results showed that eight of the genes are also present in our 124 gene list, with the same direction in expression change, three of them belonging to the group of broad categories related to cell proliferation and apoptosis. The other five genes reported by Cardoso et al. were actually not present in one of the two GEP studies mentioned in the meta-analysis. The second part of our analysis made use of freely available enrichment tools to detect which GO categories or KEGG pathways were significantly overrepresented in the three gene sets obtained from the 23 gene expression profiling studies.