The predicted values of HiSIF-DTA on the Davis, KIBA, and Human test sets

We propose a novel DTA prediction model that enables hierarchical fusion of protein semantic information to enrich protein representations and enhance DTA prediction performance. Simultaneously, we have designed two different backbones for semantic information fusion: Top-Down (TDNet) and Bottom-Up (BUNet). To validate the superiority of the model, we have trained and tested it on DTA datasets (davis and kiba) as well as the CPI dataset (Human). These files contain the predicted values of the two backbones of HiSIF-DTA on the test sets of the three datasets. 

**Please note that these predicted values are specific to the HiSIF-DTA model and the respective test datasets mentioned. **