Identification Of Linear Epitope With Multiple Sequence Representation Learning Network

Bioinformatics is an emerging field of science that integrates a variety of traditional scientific fields.Driven by developments in genetics,molecular biology and biotechnology,many disciplines,including biology,computers and information technology,have come together to organize and store vast amounts of biological information.Immunology also plays an important role in this field.b cells are the main component of the adaptive immune system,providing long-term protection against pathogens and harmful molecules.b cell epitopes are certain regions on the surface of an antigen that can be recognized by b cell receptors or antibodies and lead to a systemic immune response.Identification of Bcell epitopes is important for medical,immunological and biological applications such as disease control,diagnosis and vaccine development.Several experimental methods have been developed to identify epitopes,such as protein crystallography and nuclear magnetic resonance techniques,but these methods are usually expensive and take a long time.Therefore,several machine learning-based methods have been developed for B-cell epitope prediction.In the past decade,a series of computational methods based on machine learning and B-cell epitope data have been developed to identify B-cell epitope data based on amino acid sequences and to try to understand the biological mechanisms and meanings hidden behind the B-cell epitope data.Early computational methods were usually based on the physicochemical properties of amino acids,and then the amino acid sequences were analyzed locally using different sliding windows to obtain the physicochemical properties of each window itself and to provide a basis for determining whether it is a B-cell epitope or not.Recently,with the continuous development of deep learning techniques in machine learning,deep learning techniques have been gradually applied to B-cell epitope prediction tasks and have made some progress.However,these machine learning algorithm-based models have several limitations in identifying B-cell epitopes.First,the expensive acquisition of correct samples of B-cell epitope data increases the risk of overfitting during training,which may lead to weak generalization of deep learning-based B-cell epitope prediction models.Also,the potential of deep learning-based models seems to be underutilized,resulting in less than perfect performance.Therefore,there is a need to develop an alternative,more accurate and robust method to identify B-cell epitope prediction models,which may help in the formulation of peptide vaccines.This paper proposes a deep learning model for B-cell epitope prediction machine learning model named Epitope MSRLN,which is specified as follows:(1)To convert the Bcell epitope amino acid sequence data into data that can be processed properly by deep learning,this paper employs a model that includes Onehot,PAM250,BLOSUM62 and Prot Vec,which are used to organize and mine B-cell epitope data from different dimensions.(2)Study the design of B-cell epitope prediction model based on deep learning model.Deep learning models are composed of multiple layers of neurons with nonlinear activation and have been successfully applied to various fields including image processing and text classification.It is essential to investigate deep learning model design for B-cell epitope prediction.To be able to integrate different vector data,a CNN-based network framework is constructed in this paper to explore different potential feature spaces.(3)these four different vector data were integrated to form an ensemble model framework to improve its B-cell epitope prediction capability.To demonstrate the performance of Epitope MSRLN,we compare it with several state-of-the-art methods and test it on multiple datasets.Experimental analysis shows that our proposed method can improve Bcell epitope prediction and provide data-derived biological insights for biologists and immunologists conducting experiments.