# The Role of Nut Sensitization in Pru p 3-Sensitized Patients: A XGBoost and Generalized Linear Model Application

## TL;DR

This study uses machine learning to analyze how sensitization to Pru p 3 and other allergens affects allergy symptoms and severity in patients.

## Contribution

The novel use of XGBoost and GLM to identify predictors of clinical reactivity and severity in Pru p 3-sensitized patients.

## Key findings

- Walnut sensitization is the strongest predictor of clinical severity in Pru p 3-sensitized patients.
- Peanut and hazelnut sensitization is most common in patients with nut allergy but peach-asymptomatic.
- Machine learning models can refine risk stratification and personalize management in nsLTP-related allergies.

## Abstract

Sensitization to non-specific lipid transfer proteins (nsLTPs) is highly prevalent in Mediterranean countries. Pru p 3 from peach is a major allergen responsible for IgE-mediated food allergies. As a panallergen, Pru p 3 shows high sequence homology with nsLTPs from other Rosaceae fruits but also from botanically unrelated sources, including nuts and pollens, leading to extensive cross-reactivity complicating diagnosis and management. Given the worldwide prevalence of peanut and tree nut allergies, this study aimed to investigate sensitization patterns in Pru p 3-sensitized patients with tree nut allergy, using artificial intelligence (AI) to identify predictors of clinical reactivity and severity. Data from Pru p 3–sensitized patients with symptoms to peach and/or nuts were analyzed. Sensitization profiles were modeled using an XGBoost algorithm to explore associations with symptoms and severity. Patients sensitized to Pru p 3 and symptomatic for peach and nuts showed predominant sensitization to peanut and hazelnut, but AI revealed stronger associations between clinical reactivity and sensitization to hazelnut, walnut, and almond. Among patients with nut allergy and peach-asymptomatic, peanut and hazelnut sensitization were most frequent, while peach-symptomatic ones, walnut and almond sensitization predominated. Overall, walnut sensitization emerged as the main predictor of clinical severity and increasing number of sensitizations correlated with higher severity. The XGBoost algorithm identified specific allergen combinations associated with symptoms and severity, highlighting walnut sensitization as the strongest severity predictor. Machine learning approaches represent a promising tool for refining risk stratification and personalizing management in nsLTP-related food allergy.