Alzheimer's disease is a complex, irreversible genetic disease affecting the nerv-ous system, with an incubation period of up to 20 years. Identifying biomarkers for Alzheimer's disease is crucial for early diagnosis, thereby facilitating more effective therapeutic interventions. Many methods have been developed to identify biomarkers of Alzheimer's disease. However, these methods have many weaknesses, such as inability to handle massive data and inability to take into account complex RNA regulatory relationships. In this paper, a causal inference based feature selection method is proposed to identify Alzheimer's disease biomarkers. Firstly, the mRNA regulatory network is regarded as the causal graph. Secondly, calculating the generalized propensity score of mRNA via regulatory network. Finally, the generalized propensity score is employed to estimate the causal effect of mRNA. The key idea of this method is that utilizing generalized propensity score control for the effects of other mRNAs, thereby finding causal relationships between mRNAs and disease. The experimental results demonstrate that the generalized propensity score enables control over the effect of regulatory mRNAs and estimates the causal relationship between mRNA and Alzheimer's disease. Our method performed better than other methods. This method provides new insights into identifying disease biomarkers.