Hints from Information Theory for Analyzing Dynamic and High-Dimensional Biological Data

Abstract

Advances in biological sciences resulted in a data deluge, especially as for gene, protein, and metabolite expression. The issue of computational power needed to analyze such massive datasets is much less critical than the result interpretation task. This work deals with the latter, proposing a soft, data-driven approach, based on simple information theory concepts, as applied to classical multidimensional statistical methods. The proposed approach allows for a strong interaction between the interpretative and computational aspects of the problem fostering interdisciplinarity. The application of these methods on transcriptome data relative to immune response and cellular development reveals insightful regulations, not only on the key instructive local processes but also on the subtle, yet robust, global-scale behavior. Furthermore, these techniques are swift in utility, as no detailed a priori knowledge of the biological system in study is required, and avoid “biased” expression cutoffs that are usually required for traditional/reductionist approaches.