Validation of bioinformatic approaches for predicting allergen cross reactivity

Part of the allergenicity assessment of newly expressed proteins in genetically engineered food crops involves an assessment of potential cross-reactivity with known allergens. Bioinformatic approaches are used to evaluate the amino acid sequence identity or similarity between newly expressed proteins and the sequences of known allergens. To be useful, such approaches must be sensitive to detecting cross-reactive potential, but also capable of excluding low-risk sequences. One difficulty in comparing the effectiveness of different bioinformatic approaches has been the lack of a standardized validation and evaluation method. Here, we propose a standardized method for evaluating the sensitivity of different bioinformatic algorithms using a comprehensive database of known allergen sequences. We combine this with a previously described method for evaluating selectivity using sequences from a crop not known to commonly cause food allergy (e.g. maize) to compare the standard ">35% identity-criterion over sliding-window of ≥80 amino acids" bioinformatic approach with the previously described "one-to-one (1:1) FASTA" similarity approach using an E-value threshold of 1E-9. Results confirm the superiority of the 1:1 FASTA approach for selectively detecting cross-reactive allergens. The validation methods described here can be applied to other algorithms to select even better fit-for-purpose approaches for evaluating cross-reactive risk.