Using Deep Learning (DL) to predict Drug-Target Affinity (DTA) accelerates the process of Drug discovery. However, most of the current studies rely too much on the word embedding feature of the target sequence, ignoring the value of the prior feature. Furthermore, target sequences naturally have a spatial structure, and using only the sequences as input will ignore the Long-Range Dependencies between amino acids, thus limiting the accurate prediction of drug-target affinity. To alleviate the above problems, we proposed a Drug-Target Affinity prediction method that fuses prior features and long-range dependency sequence features (FPLRDGraph-DTA). The method utilizes a priori features of the target sequence and incorporates the long-term dependence of the embedding features. In addition, we use the Protein Language Model (PLM) to represent protein sequences, and apply it directly to the DTA prediction task as a representation of priori features, enabling knowledge transfer from sequence to structure. We compare with state-of-the-art prediction models on two benchmark datasets, and FPLRDGraph-DTA significantly improves the prediction ability of the model. The synthesized results of experiment show that introducing multiple prior features and mining Long-Range Dependencies of data play an important role in im-proving prediction performance.