In this talk, we formulate network construction as a model selection problem under the regularized sparse regression framework.It is well known that, commonly used criteria such as BIC or cross-validation for model tuning tend to select many false edges, We employ the idea of data perturbation and model aggregation and propose a re-sampling based approach which allows for estimating the false positive rate and power.As a consequence, the resulting procedure is able to control FDR at a pre-specified level and at the same time, it is less conservative compared to some existing methods.This approach can be applied to a broad range of models whenever structure learning is of interest.The proposed method is illustrated by both simulated and real data examples.