Comparative assessment of multiple criteria for the in silico prediction of cross-reactivity of proteins to known allergens

Abstract

Genetically modified crops are becoming important components of a sustainable food supply and must be brought to market efficiently while also safeguarding the public from cross-reactivity of novel proteins to known allergens. Bioinformatic assessments can help to identify proteins warranting further experimental checks for cross-reactivity. This study is a large-scale in silico evaluation of assessment criteria, including searches for: alignments between a query and an allergen having ⩾35% identity over a length ⩾80; any sequence (of some minimum length) found in both a query and an allergen; any alignment between a query and an allergen with an E-value below some threshold. The criteria and an allergen database (AllergenOnline) are used to assess 27,243 Viridiplantae proteins for potential allergenicity. (A protein is classed as a “real allergen” if it exceeds a test-specific level of identity to an AllergenOnline entry; assessment of real allergens in the query set is against a reduced database from which the identifying allergen has been removed.) Each criterion’s ability to minimize false positives without increasing false negative levels of current methods is determined. At best, the data show a reduction in false positives to ∼6% (from ∼10% under current methods) without any increase in false negatives.