Abstract: Under the framework of transfer learning, this paper further investigates high-dimensional generalized linear regression problems using elastic net penalty. To address the common challenge of insuffcient sample size in target high-dimensional datasets, we enhance estimation and prediction accuracy by leveraging potentially related auxiliary source datasets. When transferable sources are known, a novel multi-source two-step transfer learning algorithm named Trans-DEN is proposed. Numerical experiments demonstrate that the proposed algorithm exhibits enhanced effcacy and robustness in high-dimensional generalized linear regression settings with highly correlated covariates. Therefore, the proposed algorithm exhibits excellent statistical performance.