Comparing early and late data fusion methods for gene expression prediction

The most basic molecular mechanism enabling a living cell to dynamically adapt to variation occurring in its intra and extracellular environment is constituted by its ability to regulate the expression of many of its genes. At biomolecular level, this ability is mainly due to interactions occurring between regulatory motifs located in the core promoter regions and the transcription factors. A crucial question investigated by recently published works is if, and at what extent, the transcription patterns of large sets of genes can be predicted using only information encoded in the promoter regions. Even if encouraging results were obtained in gene expression patterns prediction experiments the assumption that all the signals required for the regulation of gene expression are contained in the gene promoter regions is an oversimplification as pointed out by recent findings demonstrating the existence of many regulatory levels involved in the fine modulation of gene transcription levels. In this contribution, we investigate the potential improvement in gene expression prediction performances achievable by using early and late data integration methods in order to provide a complete overview of the capabilities of data fusion approaches in a problem that can be annoverated among the most difficult in modern bioinformatics.