Research on Prediction Algorithm of Active Sites of BRET Protein Based on Deep Learning

As an important executor of life activities, the accurate identification of protein's active sites is of great significance for understanding biological functions, drug design and protein Project. However, the traditional BRET (Bioluminescence Resonance Energy Transfer) technology faces the challenge of accuracy and efficiency when dealing with large-scale data. In response to this issue, this study designed a hybrid deep learning model, BRET SiteNet, which combines Convolutional Neural Networks (CNN) and Long Short Term Memory (LSTM), to efficiently predict the active sites of BRET proteins. In this study, BRET protein samples containing 100 marked active site positions were collected and pretreated from the public protein database PDB. Protein sequence and three-dimensional structural features were extracted by single thermal coding and bioinformatics tools, and standardized and normalized. Subsequently, the BRET-SiteNet model is constructed, which uses CNN to capture the local features in the sequence, and handles the long-distance dependencies through the LSTM layer to improve the accuracy of prediction. The experimental results show that the BRET-SiteNet model performs well in the task of active site prediction, and its accuracy, precision, recall and F1 score are superior to other existing methods, such as support vector machine (SVM), random forest, CNN alone or LSTM model. Specifically, the accuracy of BRET-SiteNet reaches 0.900, and the F1 score is 0.877, which proves its powerful feature extraction ability and effective processing ability for complex protein sequences. The BRET-SiteNet model developed in this study provides a new and efficient solution for the prediction of active sites of BRET protein, which is expected to promote the development of functional research and drug design in protein.