Predicting flexible length linear B-cell epitopes using pairwise sequence similarity

Recognizing linear B-cell epitopes is important for the epitope-based vaccine design, and it have been attracting worldwide researchers. Compared to traditional experimental techniques, the computational methods for epitope prediction are faster and more economical. The earliest computational methods for linear B-cell epitopes prediction were based on some amino acid property, and their performances were poor. Recently, machine learning methods were applied to epitopes prediction as so to make improvement, and the machine learning methods usually requires the fixed-length inputs. However, linear B-cell epitopes are of varied lengths, and have to be trimmed or extended to a specific length; therefore the models based on these modified peptides can only predict the specified-length epitopes. In this paper, we developed a method named BPairwise to predict flexible length linear B-cell epitopes. First of all, we adopted an encoding scheme based on pairwise sequence similarity, which can transform the flexible-length peptides into fixed-length feature vectors. Thus, support vector machine (SVM) was used as the classification engine to construct prediction models. When applied to benchmark datasets, our proposed method can give out better results over benchmark methods in terms of accuracy, Matthew's correlation coefficient, and area under ROC curve. In conclusion, BPairwise is a tool of potential for epitope prediction.