Microarray Data Analysis: Gene Regulatory Networks

Cellular processes involve million of molecules playing a coherent role in the exchange of matter, energy and information both among themselves and with the environment. These processes are regulated by proteins, whose expression is controlled by a tight network of interactions between genes, proteins and other molecules. There is evidence that some pathologies of major social impact, like cancer and diabetes, involves groups of genes and proteins functionally related (pathways) rather than the expression of a single gene or protein. Therefore, it is important to investigate the global modifications of a specific regulatory pathway rather than the expression of a single gene. This is a major goal of systems biology approaches, devoted to the elucidation of the complex network of interacting DNAs sequences, RNAs and proteins regulating and controlling gene expression. Today, high-throughput technologies such as microarray and mass spectrometry, measure the cellular molecular expression in a given instant, thus making possible, at least in principle, the reconstruction of the regulatory network from its observed output through reverse engineering approaches. Unfortunately, microarray technology cost restricts the number of samples (order of 101-102) available for each experiment with respect to the number of monitored genes (order of 104). Mass spectrometry techniques have some limitations as well, since at present they are not able to provide a precise quantification of protein expression and require to identify the original protein from the spectrum of its fragments by mining specific databases. For these reasons, reverse engineering approaches are usually limited to very general and abstract models, where RNA expression is considered as a proxy of protein expression in controlling gene transcription. Several reverse engineering approaches based on this model have been proposed in the last years to infer gene regulatory networks from microarray gene expression data. Among them Boolean models, models based on differential equations, Bayesian networks and methods based on pair-wise gene expression correlation.