The main challenge addressed in this PhD is the detection of changes in a set of significant predictors across different contexts, applied to the differential analysis of gene regulatory networks. We propose to investigate this differential mining task by adopting a subgroup discovery view in two directions: a subspace clustering approach and a pattern mining approach. To ensure efficiency, we propose to interleave the subspace clustering and/or pattern mining steps with steps of predictor selection. The main data mining difficulty is the creation of such an interleaved process, from the study of its formal properties to the design of efficient algorithmic solutions.