Prediction Of Linear B Cell Epitopes Based On Feature Selection

The interrelationship between antigen-antibody is a key event in humoral immune response to invading pathogen,they are combined and aligned with the antigen determinant at the specific site.B-Cell epitope is one of the antigen determinant factors,and receptor(membrane bound antibody)could identify and combine it on B lymphocyte surface.B-Cell epitope,is a few amino acid fragments on the surface of the antigen.Epitopes can induce an immune response,and it is also known as antigen determining factor,and can be identified specific antibody.Generally,B cell epitope have two categories based on its structure feature:linear B cell epitope and conformational epitope.Linear B cell epitope plays an important role in biochemistry,virology,immunology,and vaccine research.Therefore,calculation method for the research and progress is still a big challenge for linear epitope prediction in bioinformatics and computational biology.Feature selection is a subset of the features and it can satisfy a certain evaluation criterion and make it achieving optimal features.Its purpose is to make the classification performance comparable or superior to that of the selected subset according to the selected subset.After a feature selection,we will eliminated some irrelevant or redundant features according to machine learning methods,thus,a more accurately model is established by using the feature subset that excluded redundant features.However,There have been a challenge on how to use feature selection method to eliminate useless or redundant features.Up to present,the most commonly used methods of linear B-Cell epitope prediction are only for the features of the extraction of B-Cells,and the existence of redundant features is not considered.In this paper,we proposed a method of combination the feature selection method with the linear B cell epitope prediction,which can eliminate the extraneous features and improve the prediction accuracy.This paper also state that different feature extraction methods corresponding to the optimal feature selection method is different by support vector machine classifier experiment,each feature extraction method corresponding prediction performance is not the same.