Evaluation and Comparison of Newly Built Linear B-Cell Epitope Prediction Software from a Users' Perspective

Background: Increasing numbers of researchers apply linear B-cell epitope prediction in their research, as in the case of peptide-based vaccine design, diagnostic tests, disease prevention and antibody production. Online software offers major ways of epitope prediction. Objective: With the advent of more and more newly built software, a standard assessment for various prediction tools is urgently needed. Methods: From the users' perspective, we compared and evaluated the ability to correctly predict true epitopes of six different B-cell epitope prediction softwares: Bepipred, ABCpred, AApred, LBtope, BEST and SVMTrip, together with one Random group and five consensus groups. Fourteen experimentally confirmed proteins (including 60 linear epitopes) were collected into an epitope database. Positive value (PV) and minor predicted value (MPV) were used to quantify the method's prediction ability. Results: Bepipred, AApred, BEST and LBtope performed significantly better than the Random group in terms of PV and MPV. Based upon the average value of the two parameters, BEST was the most efficient tool. Conclusion: Bepipred, AApred and BEST are the most efficient tools for users to predict linear B-cell epitopes. In addition, consensus groups have the effect of gathering true predicted results, but they fail to greatly improve prediction performance.