PDR: a language model for predicting protein-DNA binding sites

The interaction between proteins - DNA plays a crucial role, cell signal transduction, and metabolic processes. Accurate prediction of protein-DNA binding sites is essential for drug development. To date, various machine learning-based computational tools have been developed to identify protein-DNA binding sites. However, existing methods face challenges such as complex feature extraction steps and limited applicability. The goal of this study is to develop an efficient computational method for predicting DNA binding sites based on protein sequences. We propose a neural network framework based on protein language models (PLM) to learn feature representations of protein sequences and predict protein-DNA binding sites. The PDR predictor was trained on a dataset of protein-DNA binding sites, and its performance and generalization ability were evaluated experimentally. The results show that the PDR predictor achieves a ROC AUC of 0.8251 on the benchmark dataset, which is better than. The predictor has been experimentally validated on multiple datasets, and the results demonstrate that it outperforms traditional methods, providing a new tool for biological research.