Abstract: The study of DNA-binding proteins has important significance and role in the field of biopharmaceuticals and clinical testing. Deep learning methods have significantly improved prediction accuracy, but have encountered bottlenecks in exploiting protein structure and evolutionary information. Therefore, this paper proposes a DNA-binding protein recognition method based on dual-channel hybrid graph neural network, which uses sequence alignment to find sequence evolution information, fuses graph attention network and graph isomorphic neural network, mines the key information of DNA-binding proteins contained in protein contact map and sequence evolution, and obtains high-precision protein representation. Experimental results show that the average accuracy of this method is improved by 9.49% compared with the average accuracy of the six typical methods on the independent test set.