HELP: B-cell recognition

BCEPS implements a predictor model based on support vector machine (SVM) trained on 555 linearized conformational epitopes extracted from antibody-antigen protein structure. It also include other models based on neural network (NN) and K-nearest neighbors' algorithm (KNN), both trained on this same dataset. Extra predictions can validate true positive results.
BCEPS returns all possible n-mer peptides (n = 7 - 25) encompassed by the protein, according to the selected size. Likewise, user can also choose the threshold to consider a n-mer as an epitope. The number of peptides listed in the output can be limited and graphics can be displayed.

HELP: Select Model

BCEPS implements a predictor model based on support vector machine (SVM) trained on 555 linearized conformational epitopes extracted from antibody-antigen protein structure translated into amino acid composition. On 10-fold cross validation this model distinguishes B-cell epitopes extracted from the same antigen with an accuracy of 75.38 % ± 5.02. Moreover, in an independent dataset consisting of linearized B-cell epitopes extracted from IEDB database, this model archived an accuracy of 67.05 %. Other models based on NN and KNN trained on the same dataset translated into amino acid composition and combination of amino acid and dipeptide composition respectively, were also implemented. On 10-fold cross validation experiments, NN and KNN models archived an accuracy of 73.87 % ± 5.11 and 72.15 % ± 4.96 respectively. Extra predictions can validate true positive results.

BCEPS returns all possible n-mer peptides (n =7 - 25) encompassed by the protein. Peptides length can be select by the user in the "Size" option box. Likewise, user can also choose the threshold to consider a n-mer epitope in the "Threshold" option box (0.3, 0.4, 0.5, 0.6 and 0.7).

The number of epitopes listed in the output can be limited using "Return only B-cell epitopes", returning only the predicted epitopes when this option is selected. Moreover, when "Extended B-cell epitopes" option is checked, those consecutives peptides that are predicted as B-cell epitopes, are collapse and reported as a single epitope. Furthermore, when "Collapse B-cell Epitopes" is selected, BCEPS computes the prediction by residue. BCEPS calculate the mean probability of being a B-cell epitope from all the peptides (of length n, selected in the "Size option") that include the residue. Then, sequential residues predicted to be B-cell epitopes are joined and BCEPS returns those with a length equal or higher than the size established by the user.

BCEPS displays a table reporting by default the start position of the peptides, the sequence of the peptides and the prediction performed by the model selected previously by the user (1/0). Here follows two representative outputs when all "Epitope features" options are selected (see in the next section). First one (A) represents the output when only "Display graphics" is selected, while second (B) represent the output when "Return only B-cell epitopes" and "Extended B-cell epitopes" options are also selected.

In BCEPS, users can also select to retrieve different physico-chemical features including ectodomain location, glycosylation, flexibility, accessibility and hydrophobicity. Results can be sort and filtered by this physico-chemical properties. As well, HLA-DRB1 binding profile and population coverage for each epitope can be reported and final output can be filtered according to the resulting profile.

HLA-DRB1 binding profile and population coverage for each epitope is reported when "CD4 T cell Epitope" option is selected. Users can select between all humans or mouse MHC II alleles or a specific MHC II allele. Final output can also be sorted according to the population coverage.