Evaluation of In Silico Prediction Possibility of Epitope Sequences Using Experimental Data Concerning Allergenic Food Proteins Summarised in BIOPEP Database

The aim of the study was to evaluate the possibility of predicting potential epitope sequences and location in allergenic proteins from food using EVALLER program by comparison with experimental epitopes summarised in the BIOPEP database of allergenic proteins. Sequences of experimental epitopes from food allergens, present in the BIOPEP database of allergenic proteins were used in the study. Sequences of potential epitopes were found using EVALLER program. The Positive Predictive Value (PPV) has been used as a measure of prediction quality. The potential epitopes fully or partially overlapping with the experimental ones were considered as true positive results whereas these unrelated to the experimental ones as false positive results. The PPV for entire dataset containing 310 potential epitopes was 80.6%. The PPV varied significantly among particular allergen families defined according to the AllFam database. Caseins revealed PPV=100% (with one exception), proteins from tropomyosin family and proteins from papain-like cystein protease family - exceeding 50%. The last two families possess also relatively low frequency of epitope occurrence. The predictive potential was poor (less than 50%) for plant allergens from cupin superfamily. Families such as lipocalins from milk and EF-hand family (parvalbumins) revealed high variability within family. The EVALLER program may be used as a tool for the prediction of epitope location although its potential varies considerably among allergen families. High PPV is associated with a high number of known experimental epitopes (such as in caseins) and/or a high degree of sequence conservation within family (caseins, tropomyosins).