Content based search in gene expression databases

The problem of content-based searching of time-series microarray experiments in large-scale gene expression databases, for the first time, is investigated in this study. The problem is examined as an information retrieval task where an entire experiment is taken as the query and searched through a collection of previous experiments. The relevant experiments are required to be retrieved based on the content similarity rather than their meta-data descriptions. A comparison of different fingerprinting and distance computation schemes is presented over a retrieval framework based on the differential expression of genes in varying time points. For all experiments carried out on database we create, results show that Pearson Correlation Coefficent and Tanimoto Distance present about 15% better performance than Euclidean Distance in comparison fingerprints based on differential expression.