| Phage display is a multifunctional panning technology,which can select peptides or antibodies that bind to specific targets.Researchers in various fields use a variety of targets to pan phage display random peptides libraries,which can usually obtain peptides bound to the target.Antibody is one of the most commonly used target.The polypeptide binding to the antibody variable region is not only helpful to determine the epitope recognized by the antibody,but also can be used in the diagnosis and treatment of related diseases.However,when using antibodies as peptide phage display panning targets,the general purpose is to screen peptides bound to antibody variable regions.However,in the experimental results,there are many peptides that may bind to regions outside the variable regions of the antibody,such as bind to the Fc fragment of the antibody.At this time,Fc binding peptide is a target-unrelated peptide(TUP).Of course,if the purpose of the study is to analyze the interaction between Fc fragment and Fc receptor or obtain Fc binding peptide for the development of affinity purified ligand of antibody,in this case,Fc binding peptide is the signal we want.In either case,it is very meaningful to determine the Fc binding peptide.At present,there is no bioinformatics study to predict the binding peptide of Fc fragment.In this study,the previously unrecorded phage random peptide panning data were collected from the published literature to update the BDB database.Then 46 Fc binding peptides,232 Fab binding peptides,9063 antibody binding peptides and 20270 non antibody binding peptides were collected from BDB database.46 Fc binding peptides and 232 Fab binding peptides were used as positive and negative samples of the training set respectively.17 feature extraction methods provided by i Feature were used to convert the sequence into numerical feature information,and further use MRMD2.0 tool to reduce the dimension of the feature matrix.Support vector machine algorithm(SVM)is used to model the features after dimensionality reduction,and four sub models are constructed.The sub models are evaluated by 5-fold cross validation.the sub models are integrated by means of mean integration strategy,and the integrated model Fc Binder is constructed.In addition,other machine learning algorithms such as Naive Bayes,Logistic,Decision Tree,KNN and Random Forest are used to establish the prediction model for systematic comparison with Fc Binder.The reliability of the model was evaluated by using 9063 antibody binding peptides and 48 Fab binding peptides from biological panning as the test set;Then,the polypeptide sequences from abiotic panning sources that can bind to Fc fragment were used to evaluate the generalization ability of the model.Finally,the Fc Binder integrated prediction model constructed by SVM has the best effect.The prediction accuracy of Fc Binder is 84.8%,and the AUC is 0.988.In addition,the integration model Fc Binder is constructed using different negative samples,and five integration models are constructed through other machine learning methods.The performance of all models was evaluated using polypeptide data from biological panning sources and Sp A domain sequences.The prediction result of Fc Binder is the best.Therefore,in order to facilitate use,Fc Binder's online prediction service is provided: http://i.uestc.edu.cn/Fc Binder/?... |
PDF Full Text Request