On the Accuracies of Sequence Based Linear B Cell Epitope Predictors

Abstract

The accuracy of online tools employed in attempts to predict B-cell epitopes based on sequence are very poor. In order to improve the accuracy of these predictions it is essential to design algorithms to benefit from the features achieved in wet lab in vivo experimental models. To shed some light on accuracy and reliability of these online tools, we set an insilico experiment on five selected online tools using five antigens whose b-cell epitopes are known through wet lab experiments. To evaluate successes of online tools, we defined two measures, accuracy, and reliability. To the findings of this experiment, the most accurate tool is ABCpred with a score of 43.59 %. That is peptides that are predicted as b-epitopes, cover in average 43.59 % of the wet lab listed b-epitopes. The most reliable predictors are BCpred and AAPred with scores of 52.54%, and 52.60% respectively, which means that in average around half of the peptides that are predicted as a b-epitope by these predictors have a chance to be a real b-epitope. Combining several predictors to get better predictors is not an advisable technique. From this experiment it is concluded that the accuracy and reliability of online tools still are far away being satisfactory.