Identification of biologically significant genes from combinatorial microarray data

High-throughput microarray technology has enabled the simultaneous measurement of the abundance of tens of thousands of gene-expression levels, opening up a new variety of opportunities in both basic and applied biological research. In the wealth of genomic data produced so far, the analysis of massive volume of data sets has become a challenging part of this innovative approach. In this study, a series of microarray experimental data from Yersinia pestis (Y. pestis), the etiologic agent of plague in humans, were analyzed to investigate the effect of the treatments with quorum-sensing signal molecules (autoinducer-2 and acyl-homoserine lactones) and mutation (ΔypeIR, ΔyspIR, and ΔluxS) on the variation of gene-expression levels. The gene-expression data have been systematically analyzed to find potentially important genes for vaccine development by means of a coordinated use of statistical learning algorithms, that is, principal component analysis (PCA), gene shaving (GS), and self-organizing map (SOM). The results achieved from the respective methods, the lists of genes identified as differentially expressed, were combined to minimize the risk that might arise when using a single method. The commonly detected genes from multiple data mining methods, which reflect the linear/nonlinear dimensionality and similarity measure in gene-expression space, were taken into account as the most significant group. In conclusion, tens of potentially biologically significant genes were identified out of over 4000 genes of Y. pestis. The "active" genes discovered in this study will provide information on bacterial genetic targets important for the development of novel vaccines.