Research On Prediction Of Nucleic Acid Binding Proteins Based On Deep Neural Network

With the development of next-generation sequencing technology,protein sequences are increasing rapidly.Protein is very important for life activities,such as gene transcription,regulation and translation.Among these important proteins,there is a kind of RNA / DNA binding protein which can interact with RNA / DNA.They are collectively called nucleic acid binding proteins.And they play an important role in the research of disease generation and drug target design.Unfortunately,up to now,the mechanism of the interaction between protein and nucleic acid has not been completely determined,and the traditional biological experimental methods need a lot of manpower,money and time,which can not meet the needs of large-scale protein data testing.Therefore,how to predict proteins that can bind to RNA / DNA from massive protein data by computational methods is an important topic in bioinformatics.In view of the importance of DNA/RNA binding proteins in biological field,this paper studies DNA and RNA binding proteins based on deep neural network.The main work includes the following two aspects:1.Prediction of RNA Binding Proteins.RNA binding proteins(RBPs)are widely involved in various biological processes,including gene regulation and protein synthesis.It is important to recognize RBP from non RBP for understanding these processes.Although there are many calculation methods for RBPs prediction,the prediction accuracy of the predictor is still not ideal.In this paper,a new method,DeepMVF-RBP,is developed.It extracts protein features from different perspectives,and then sends the multi-view protein features into the deep belief network(DBN)to automatically extract high-dimensional abstract features,and then predict whether the protein is RBP or not.Multi-view feature representation can not only capture local sequence and global sequence information,but also sequence-order information.This method is trained and tested on two datasets.The results show that the importance of multi-view feature representation learning in predicting RNA-binding proteins and the model constructed in this paper is a powerful theoretical framework for studying RNA-binding proteins.2.Prediction of DNA Binding Proteins.DNA binding proteins play an important role in many biological processes,such as specific nucleotide sequence recognition,transcription and DNA replication.In addition,DNA binding proteins have many important applications in the treatment of genetic diseases,DNA biology research and drug development.In this paper,a new prediction method,MsDBP,is proposed,which is a deep neural network prediction method combining the characteristics of multi-scale sequences.Instead of describing the whole protein directly,this paper divides the whole protein sequence into different length sub-sequences,and then encodes them according to the composition information.In this way,we can get the multi-scale sequence characteristics of this protein.Finally,the integrated dense layer is used to learn abstract features to identify DNA binding proteins.The prediction results of three independent test sets and one large-scale DNA binding protein set show that the proposed method is feasible and effective.In addition,this paper also built the corresponding online prediction website.In short,in the prediction of RNA-binding proteins,the method focuses on multi-view feature representation of protein sequences,and combines deep belief networks to make up for the shortcomings of traditional machine learning methods which are not good at extracting hidden association between features.In the prediction of DNA-binding proteins,we focus on extracting multi-scale sequence characteristics of proteins,and analyze the impact of 10 DNA-binding proteins on human life activities.