What do the scores on the Y-axis mean. How does one set a threshold or cut-off?
For each input sequence, BepiPred outputs a prediction score for every residue. The positions of the linear B-cell epitopes are predicted to be located at the residues with the highest scores.
The expected sensitivity/specificity of the prediction is a function of the score cutoff or threshold the user chooses. There is a table on: http://www.cbs.dtu.dk/services/BepiPred/output.php where this is shown. It is based on the prediction of epitopes and non-epitopes. Sensitivity, or true positive rate, is the proportion of correctly predicted epitope residues (TP) with respect to the total number of epitope residues (TP + FP) or TP / (TP + FP), where TP = # of true positive and FP = # of false positive. Specificity is the proportion of correctly predicted non-eptiope residues (TN) with respect to the total number of non-epitope residues (TN+FP). It equals 1 - false positive rate = 1 - FP/(TN+FP), where FP = # of false positive and TN = # of true negative. As an example, for scores above 0.9 the sensitivity is 0.25 and the specificity is 0.91. The user can vary the threshold score to trade off between the sensitivity and specificity. The paper by Larsen, Lund, and Nielsen describing BepiPred (PMID: 16635264) can be downloaded from BioMed Central or the IEDB website at http://tools.iedb.org/main/pdfs/16635264.pdf.
Ward Fleri, PhD
IEDB Project Manager