Combination methods in microarray analysis

Microarray technology and experiment can produce thousands or tens of thousands of gene expression measurements in a single cellular mRNA sample. Selecting a list of informative differential genes from these measurement data has been the central problem for microarray analysis. Many methods to identify informative genes have been proposed in the past. However, due to the complexity of biological systems, each proposed method seems to perform nicely in a particular data set or specific experiment. It remains a great challenge to come up with a selection method for a wider spectrum of experiments and a broader variety of data sets. In this paper, we take the approach of method combination using data fusion and rank-score graph which have been used successfully in other application domains such as information retrieval, pattern recognition and tracking, and molecular similarity search. Our method combination is efficient and flexible and can be extended to become a general learning system for microarray gene expression analysis.