Predicting an individual’s disease risk from the genotype data is a central problem in human genetics; due to modest effect sizes of each variant, this problem is still challenging. Recently, deep learning methods have emerged as powerful pattern-recognition techniques and have been widely used in genetics and genomics. However, applying standard deep learning methods alone are unable to improve disease risk prediction accuracy in general compared with other competing methods (Bellot et al. Genetics, Vol. 210, 809–819, 2018). In this paper, we propose a new deep learning framework that integrates regulatory and transcriptome information with genotype data. Specifically, we leverage gene expression imputation from genetic data and then train a deep learning model based on the imputed gene expression to predict disease risk. We will apply our method to UK Biobank dataset, and hopefully, we could showcase the power of deep learning methods and gain new insights into the genetic basis of complex diseases.