Some Extensions For The Approaches Of Microarray Data Meta-analysis

A primary goal of microarray experiments is to identify differentially expressed genes (DEGs) under different biological conditions. A major problem met in microarray studies is that the reproducibility of results between individual microarray experiments is usually poor. A main cause for this situation is that the sample size (the number of replicate microarrays) in microarray experiments is usually small. Meta-analysis appears to be a promising and practical solution for this conundrum. In this study, three extensions in the methodology of microarray data meta-analysis were investigated:1) the very popular microarray analysis software SAM was applied to meta-analysis;2) meta-analysis was performed on the microarray data from opposite physiological processes; and 3) meta-analysis was performed on the microarray data from related experiments without replicates. The main results are as follows:1. In the first practical example, microarray data were from four independent cold stress (4℃,24h) experiments in Arabidopsis. The DEG lists obtained by separately analyzing in individual experiments using SAM were very different. Among the-13000 genes analyzed in total, only 317 up-regulated genes (URGs) and 132 down-regulated genes (DRGs) were simultaneously detected in all the four experiments; whereas,3134 URGs and 2983 DRGs were identified by the meta-analysis using SAM. Comparison indicated that over 80% of the DEGs simultaneously significant in at least two experiments were also detected by the meta-analysis. It is proved by GO and promoter motif analyses that the DEGs identified by the meta-analysis with SAM were related to cold stress. Therefore, the application of meta-analysis with SAM is feasible.2. The second practical example consisted of a drought stress experiment using 4 microarrays and a rehydration experiment using 2 microarrays (each array containing-24132 genes) in Arabidopsis. SAM and a professional microarray meta-analysis software RankProd were used for detection of DEGs. In the drought stress experiment, 1860 and 1188 DEGs were identified by SAM and RankProd, respectively. Considering that drought and rehydration were two opposite physiological processes, the rehydration data were merged with the drought data after multiplied by (-1) for meta-analysis. There were 2306 and 1978 DEGs identified by SAM and RankProd from the merged data, respectively. It is indicated by comparison that most of the DEGs detected by the separate analysis in the drought stress experiment were also detected by the meta-analysis. It is proved by GO and promoter motif analyses that the DEGs identified by meta-analysis were related to drought stress. This result showed that meta-analysis of microarray data from opposite physiological processes is feasible, from which more and reliable DEGs could be detected. The result also indicated that SAM has a high statistical power than RankProd in meta-analysis.3. In the third practical example, microarray data were from 6 different experiments of appressorium induction in Magnaporthe grisea. Because only one microarray was used in each experiment, DEGs could be only identified with the criterion of 2-fold change. By comparison, only 67 DEGs were detected from the 6 datasets simultaneously. By performing meta-analysis with SAM,485 URGs and 457 DRGs were identified. GO analysis showed that the DEGs detected by the meta-analysis were related to the appressorium formation in Magnaporthe grisea, consistent with the results of other researchers. This result suggested that meta-analysis of microarray data from related experiments is feasible.In conclusion, the practical applications suggest that the extensions made for meta-analysis in'this study are feasible, which provide valuable ways for the reanalysis of great amount of microarray data.