Integrative construction of gene signatures based on fusion of expression and ontology information

Gene signatures are lists of genes used for summarizing high-throughput gene expression profiling experiments. Various routines for obtaining and analyzing gene signatures in molecular biology studies exist, including statistical testing with false discovery corrections and annotations by gene ontology keywords. Despite the presence of well established routines there are still challenges in efficient application of gene signatures, which include gene signature instability, problems in defining optimal sizes and possible unreliability of inference results. Therefore there are continuous attempts towards improving algorithms for constructing meaningful gene signatures. In this paper we are introducing a methodology for constructing gene signatures, based on the fusion of information coming from statistical tests for differential gene expression analysis and resulting from statistical tests for GO terms enrichment analysis. On the basis of the DN Amicroarray datasetswe are demonstrating that the proposed algorithm for fusion of expression and ontology information leads to improvement of the composition of gene signatures.