This thesis proposes a method to build realistic causal regulatory networks hat has lower false positive rate than traditional methods. The first contribution of this thesis is to integrate heterogeneous information from two types of network predictions to determine a causal explanation of the observed gene co-expression. The second contribution is to model this integration as a combinatorial optimization problem. We demonstrate that this problem belongs to the NP-hard complexity class. The third contribution is the proposition of a heuristic approach to have an approximate solution in a practical execution time. Our evaluation shows that the E.coli regulatory network resulting from the application of this method has a higher accuracy than the putative one built with traditional tools. The bacterium Acidithiobacillus ferrooxidans is particularly challenging for the experimental determination of its regulatory network. Using the tools we developed, we propose a putative regulatory network and analyze it to rank the relevance of central regulators. In a second part of this thesis we explore how these regulatory relationships are manifested in a case linked to human health, developing a method to complete a linked to Alzheimer 's disease network. As an addendum we address the mathematical problem of microarray probe design. We conclude that, to fully predict the hybridization dynamics, we need a modified energy function for secondary structures of surface-attached DNA molecules and propose a scheme for determining such function.
