Cluster analysis of food-allergy patient data reveals patterns of co-sensitization

RATIONALE: Patients can be allergic to multiple substances due to IgE-mediated recognition of similar epitopes on proteins from different sources. We applied cluster-detection techniques to food-allergic patient data to detect groups of cross-reactive allergens. Such groupings will be useful for patient-classification, diagnosis, treatment and discovery.

METHODS: Skin prick test (SPT) results were obtained for confirmed food-allergic patients, for allergens common to Mediterranean areas. Patients/allergens with much missing data were excluded. Cluster analysis was performed using R/Cytoscape. Similarity was calculated using binary distance metrics. Patient self-reporting data was also obtained.

RESULTS: Following exclusion, 525 participants and 45 agents were analysed. The allergens giving rise to the most positive SPT results were olive pollen, peach, tree-nuts/peanuts, grasses and house-dust mites. Cluster analysis found that similar allergen-sources tended to group together, including fruits, mites, nuts, dander, trees, weeds and grasses. Comparison with self-reported previous reactions showed high overlap, albeit with notable exceptions including lentils and sesame seeds. The choice of distance metric and clustering method influenced cluster-building.

CONCLUSIONS: SPT analysis reveals patterns of co-reactivity between allergens. This information can aid diagnosis and suggest which allergen-sources to avoid. It can also guide studies of panallergens and epitope mapping. However, the choice of metric to calculate similarity is important: given the predominance of negative data, assymetric metrics are advised. Future work will investigate other geographical areas and patient IgE levels.