Research On Protein Domain Interaction Sites Prediction

Proteins are essential components of organisms and participate in every process of cell life,and the domain is the basic unit for proteins to maintain their structural characteristics.There are usually multiple domains in a protein,and the same domain is located in different positions in different proteins.Protein interaction involves the binding of corresponding domains.Accurate recognition of domains is crucial for protein structure resolution,as well as predicting and understanding protein function.In this paper,we use the method of deep learning to study the prediction of protein domain interaction sites,and design and develop a prediction tool based on the model and software development technology.Firstly,the data sets are selected from the 3D interactive database,and the data are preprocessed.Then the PSSM,secondary structure information,relative solvent accessibility and sequence one-hot coding are used as feature inputs,and convolution neural network and bidirectional GRU network are established to learn the input data and finally a fully connected neural network is used to predict the interaction type.The results show that the Recall,MCC and F1 of the model are 0.482,0.129 and 0.321 on the Dset?164 dataset,which show that the model has better generalization ability than PSIVER and SPRINGS benchmark algorithm.Finally,for the domain interaction prediction of the Web system design and development,this paper uses B/S architecture,combined with Flask,Celery and other technologies,at the same time,the system framework design and system implementation process.Users can submit the sequence text and query and visualize the interaction sites on the Web page,which is helpful for the subsequent 3D modeling of proteins.In summary,the prediction of protein domain interaction sites is studied in this paper,and the model has good performance.At the same time,provide researchers with online forecasting tools.The purpose of this study is to provide some help for predicting residual contact at the domain level.