Prediction of Protein-DNA Binding Sites Based on Protein Language Model and Deep Learning

Proteins binding to DNA is crucial for biological processes and drug development. The current computational methods are limited by the high cost of data acquisition, complex processing process, and incomplete engineering representation of manually designed feature extraction. Therefore, based on DNA-binding protein sequence information, a feature extraction method combining manual features and pre-trained models is proposed. Secondly, deep learning methods are used to capture local sequence features and long-term dependencies within the sequence, respectively. Finally, the attention mechanism is introduced to integrate features and learn weights. The performance of the latest protein language model is compared with that of the mainstream method on the test set. The MCC value of the proposed method is improved by 22.1% on average. The comparison results prove the efficiency and accuracy of the method.