Background: In epigenome-wide association studies (EWAS), the mixed methylation expression caused by the combination of different cell types may lead the researchers to find the false methylation site related to the phenotype of interest. In order to fix this problem, researchers have proposed some non-reference methods based on sparse principle component analysis (PCA) to correct the EWAS false discovery. However, the existing model assumes that all methylation site have the same a priori probability in each PC load, but it is known that there already has network structure in the genetic variable corresponding to the methylation site. In this paper, we show that the results of the existing EWAS correction model are still not good enough. If we can integrate the existing methylation network as prior knowledge into the sparse PCA model, we can effectively improve the correction ability of the existing model. Result: Based on the above ideas, we propose GN-ReFAEWAS, a model which uses the prior methylation gene network structure into the PCA framework for feature extraction. This model can be used to correct the false discovery in EWAS. GN-ReFAEWAS model does not need cell counting data and can estimate cell type composition through methylation principal component data. The key of this model is to solve a sparse regularize problem of methylation network. This paper uses regularize and random sampling algorithm to solve this problem. We used one simulated data set and three real data sets for experiments and compared four existing EWAS calibration models. The experimental results show that the GN-ReFAEWAS model is superior to existing models. Conclusion: The result proved that GN-ReFAEWAS model can provide a better estimation of cell-type composition and reduce the false positives in EWAS.