Poster: Linear B-cell epitope prediction based on Support Vector Machine and propensity scales

Abstract

B-cell epitopes play an important role for developing synthetic peptide vaccines and inducing antibody responses. Applying biological experiments for epitope identification is time consuming and demands a lot of experimental resources. Nevertheless, it is important yet challenging task for designing a computer-aided B-cell linear epitope prediction system with high precision rates. In this paper, a combinatorial mechanism based on physico-chemical properties and SVM (Support Vector Machine) techniques for linear epitope prediction is proposed. Amino acid segments (AASs) with 2, 3 and 4 residues in length of both epitopes and non-epitopes datasets [1, 2] were trained and applied as statistical features of SVM [3]. The proposed system was evaluated by one curated dataset and two public epitope databases, and its performance was compared with four existing approaches. The experimental results have shown that our proposed method outperforms other existing systems in terms of specificity, accuracy, and positive predictive value in most testing cases. Besides, the sensitivity is also achieved with a comparable performance.