Methodology to Explore Co-expression in Microarray Data

In the past several years, the amount of microarray data accessible on the Internet has grown dramatically, representing millions of Euros worth of underused information. We propose a method to use this data in a coexpression study. The method is simple in principle: the aim is to detect which genes react in the same way in certain circumstances (such as a disease, stress, medication,), potentially highlighting new interaction partners or even new pathways. We propose to study coexpression using a large amount of data, process it through an adequate algorithm and visualize the results with a dynamic graphical representation. We gather the microarray data using the PathEx database developed in our lab, which allows searching through more than 120,000 microarrays experiments on Homo sapiens using specific criteria such as the tissue sample, the biological background or any information contained in the metadata describing the experiment. Then, we process the data using the Minet R package, which allows for coexpression analysis using cutting-edge algorithms such as ARACNE or MRNET methods. This step computes the weighted relations between all the probesets in the microarrays and provides a GraphML representation of the relations. In order to explore the relations optimally, we channel the GraphML into a dynamic graphical program we developed called gViz. This program allows for data visualization but also for exploration and post-analysis. We can extract meaningful information from the network computed, compare this information with curated databases such as KEGG (Kyoto Encyclopedia of Genes and Genomes), highlight the discrepancies and hopefully discover new interactions or add new steps in canonic pathways. We present here a fast, free and user-friendly working methodology to analyze co-expression in microarray data