Exploiting inter-gene information for microarray data integration

Microarray data integration is an important yet challenging problem. Usually, direct integration of microarrays after normalization is ineffective because of the diverse types of experiment specific variations. To address this issue, two novel integration approaches were proposed in recent microarray studies. The first study[16] presented a cancer classification technique which identifies gene pairs whose expression orders are consistent within class and different across classes. The other study[18] presented a promising gene expression analysis technique which utilizes pairwise correlations of gene expressions across different microarray datasets. Interestingly, we observe that both of the independently developed techniques rely on inter-gene information and noise filtering strategy to achieve satisfactory performance in microarray integration. Motivated by this observation, we propose in this paper a formal data model for microarray integration using inter-gene information and effective filtering, which generalizes the previous two frameworks. We also show how the proposed model can handle a broader range of problems than the previous frameworks.