A method for improving the reliability of causal inference from large-scale data in biomedicine

Causal inference is an essential problem in the field of drug discovery. Most existing methods depend on hypothesis testing with the analysis of drug-disease or drug-target-disease relations. However, a remaining challenge is how to choose a significance level for conditional independence tests to achieve the most reliable causal relation result. In this paper, we propose a constraint-based causal discovery method to achieve more reliable causal relations. An independence test reliability (ITR) is adopted to measure the reliability of each independence test result, and an aggregated causal relation reliability (ACRR) is introduced to evaluate the global reliability of a causal discovery. A drug-target-disease dataset is established by collecting information from the literature in PubMed using data mining techniques. The result of our experiments on different datasets shows that our proposed method obtains more stable causal relations than the existing approaches.