Benchmarking the propensity scales for the prediction of linear B-cell epitopes

Subunit vaccine designing is an integral part of vaccine design strategy which requires the identification of T and B-cell epitopes in an antigenic protein sequence. Although conformational aspects of antibody binding complicates the problem of experimental B-cell epitope identification, the prediction of linear B-cell epitopes is useful in the development of peptide based vaccine and immunodiagnostics. Available computational methods for prediction of linear-B cell epitope are bound to use for a fixed window size (epitope length) and they are not evaluated comprehensively on the dataset of experimental B cell epitopes.

The present work is related to analyze length of B cell epitopes found in Bcipep database which ranges from 5–20 amino acids. In this study effort has been made to benchmark propensity scales of six physico-chemical properties viz. hydrophilicity, flexibility, polarity, turns, exposed surface and antigenic propensity, which are being used to predict linear B cell epitopes in protein sequence. The prediction accuracy of the individual propensity scale for these six physico-chemical properties varies from 51.00% to 58.37%. Their performances are also evaluated by ROC curve for the thresholds ranging from −3.0 to + 3.0. Out of six physico-chemical properties, flexibility is the most accurate scale because of its maximum Aroc value (0.60). In order to improve the prediction accuracy a combined scale is used taking in to account the flexibility (Karplus et. al.) with hydrophilicity (Parker et. al.), Polarity (Grantham) and Exposed surface (Jenin) which shows better accuracy (59.74%) at a threshold 0.65 to the other possible combinations.