Computational Methods For B-cell Epitope Prediction Based On Mimotope Mapping

In the field of immunology, the prediction of B-cell epitope has beenan essential issue which is extensively discussed in recent years. B-cellepitope, which resides on a consecutive region on the surface of anantigen and can be recognized by BCR and subsequently antibody, is ofcrucial importance for elucidating the mechanism of molecularrecognition and vaccine design. However, confined by the experimentalmethods mainly crystal diffraction and NMR which is both arduous andcostly, limited epitopes are found. Therefore, computational methodsfeatured for its low cost and high speed was employed to predict B-cellepitopes in silico. In my two years’ study, the major attention was focusedon devising better algorithms based on epitope mapping.Under the force of survival, the epitope on an antigen surface, whichis recognized and bound by antibody, is less conserved than the interfacesin other types of protein-protein interactions. It is suggested that theinteracting epitope on an antigen is context dependent that an antigenusually contains several different epitopes for each antibody. To address this complexity, a combinatorial method was proposed which is based ona number of affinity-selected peptides from phage-display experiment.Since these affinity-selected peptides serve as functional equivalents tothe genuine epitope, they are commonly termed as mimotopes and used toreveal the location of the epitope on antigen surface.There are a multiple of ways to map mimtopes back to an epitope. Inthis study, we tried and alter the graph search model to attain betterperformance. With an adaptable distance threshold guided bycompactness factor, the method improved from the the original modelwhich better reflects the flexibility of the interaction between any aminoacid pairs. Beside, the method brought in the idea of patitioning the wholegraph of an antigen surface into smaller groups to reduce the complexityof graph search. On a dataset comprising17cases, the methods reachedthe highest sensitivity and precision compared with the other twostate-of-the-art methods as well as the highest speed.In a direct continuation of this work, an assembled idea wasintroduced to epitope prediction. Intuitively, combining the outcomesfrom multiple tools, the integrated tools can be used to assess theperformance of each prediction so that better performance can be expected. The methods provided the best performance on sensitivity andprecision comparing with the individual methods. The results alsodemonstrated the consistency of the results from multiple tools could be aideal indicator about the reliability of the prediction. For furtherevaluation and the use of the community, the assembled method wasdeveloped as web-based servers athttp://informatics.nenu.edu.cn/PepMapper.