Research Of B-cell Epitope Prediction Method Based On Multi-information Fusion

In recent years, the B-cell epitope prediction was discussed as a central issue, which occupies a very important position in the field of immunology. B cell epitope refers to such a group of amino acid residues, they are located on the surface of antigens, and these residues can be recognized by the B-cell surface receptor and combine with antibody specificity. It has very important significance to locate these epitopes on antigen for the purpose of high-throughput antibody preparation, immune intervention therapy and artificial vaccine design. The most reliable methods for identification of an epitope are biological experiment techniques, such as X-ray crystallography, NMR and etc. are the techniques to parsing the spatial structure of antigen-antibody complexes, thereby locating B-cell epitope accurately, but they are time consuming and expensive, also have high requirements for equipment. With the increase of known B-cell epitope data, candidate epitopes that are selected on the surface of the antibody by computational methods at first, and then identify and confirm the candidate epitopes through biological experiment technology, this kind of method has been widely used, and substantially accelerating the identifying process of B cell epitope prediction.Conformational B-cell epitope prediction methods can be divided into two categories: structure-based B-cell epitope prediction and mimotope-based B-cell epitope prediction. The two kinds of methods seems to have reached a performance bottleneck. In this paper, we proposed a new B-cell epitope prediction method based on Multi-information Fusion, we select the commonly used 3D structure information of antigen, mimotope information and antigen-antibody interaction information for fusion, then using machine learning methods to establish prediction models and algorithm for conformational B-cell epitope prediction. In this paper, we select 18 test cases as a testing datasets which have only one mimotope sets for one complex structure from the benchmark datasets of “Mimotope benchmark 2.0”. A large number of experimental proof that the parameters I=5&K=0&S=1 which gave the highest values for RF by using the testing dataset. The proposed method gives out the sensitivity of 0.60, the specificity of 0.81, the PPV of 0.35, the MCC of 0.15, the ACC of 0.80 and the mean F score of 0.44 on the testing dataset. When compared with other publicly available servers: Pep-3D-Search, Epi Search, Pepsurf and Mimopro, the new method yields better performance on the value of PPV, MCC, ACC and the F score. The results demonstrate the proposed method is correct and effective.