B-cell Epitope Prediction Using Extreme Learning Machine and Particle Swarm Optimization-based Undersampling

B-cell epitope prediction is the task of estimating the class label of antigen surface as the epitope or non-epitope. Since each protein dataset consists of different scales, such as physicochemical, statistical and structural, it may be efficient to identify the importance of each scale and its influence on the results of the prediction. To this end, this paper uses some criteria to select the most important scales of the dataset. Also, the problem of imbalance distribution of samples over epitope and non-epitope classes is solved in this paper using Particle Swarm Optimization (PSO) algorithm which searches on the samples of the majority class and remove the least significant samples from the majority class. The prediction model used in this paper is Extreme Learning Machine (ELM) that benefits from the extremely fast learning speed and appropriate generalization performance. Evaluation results of the proposed method on a protein dataset with different number of samples and scales shows the priority of the proposed method compared to the similar epitope prediction models.