Many experimental observations and known cellular mechanisms are available but large-scale analyses of such information remain difficult due to the heterogeneity of the biological mechanisms. In this context, we introduce a new automatic method to integrate this information in order to find potential key regulators of a set of molecules. First, we introduce the concept of ”regulated reaction” to gather in a unified formalism information about reactions (consumption and prediction of molecules) and causalities (effect of the variation of a molecule on the variation of another molecule). This formalism is then modeled as a causality graph describing in a predictive way consequences of the variations of fluxes and molecules quantities. Finally, we use graph-based algorithms and biologically relevant scores to extract key regulators of a set of molecules. We validate this method by recovering among the causal graph derived from the Transpath database the main regulators of 190 groups of genes which are known to share a transcription factor according to the TRED database.