Figure 2.

Overview of the NHGNN-DTA framework. The adaptive feature generator is shown in the upper part of the figure. It is designed to generate high-dimensional feature representations of amino acids and atoms. It obtains a good feature representation through a multi-head attention mechanism in the pre-training stage and then dynamically and intermittently transfers the normalized features to the HGNN. As shown in the lower part of the figure, HGNN realizes graph-level information interaction through the only central node “C.” In the joint training of the feature generator and HGNN, the features of the hybrid graph are adaptively obtained for further optimization. Finally, the predictions of the DTA are jointly output.