A core task of drug discovery study is to identify the dependency between the genetic/ molecular makeups of the human body and disease phenotype.Here we proposed a novel method for systematically screening the microRNAs regulating cancer prognosis via driving a specific ppi (protein-protein interaction) network.In this method, we integrated information from microRNAs expression, mRNAs expression and patient outcome, and tried to mine the in vivo dependency between microRNAs and ppi networks.The conditional dependency of the mRNAs expression (G), microRNAs expression (M) and phenotype (P) was calculated, and the permutation was conduct to screen the significant Conditional Mutual Information Ⅰ (G, P|M).In this study we test the proposed method on autophagy protein networks in ovarian cancer as example.Autophagy is a cellular process involved the degradation of cells own components, in which cytoplasmic contents are sequestered into a double-membrane called autophagosome and then subsequently delivered to lysosome for degradation.We identified a group of microRNAs, which could regulate the hub nodes of autophagy ppi network.We experimentally demonstrated that the transfection of the mimics of these microRNAs could inhibit autophagy via inhibiting the target gene: Beclin1.Finally, the miRNA-target gene pair is significantly associated with prognosis in a cohort of ovarian cancer patients .