Food anaphylaxis diagnostic marker compilation in machine learning design and validation.

Area of Science:

• Immunology
• Computational Biology
• Pediatrics

Background:

• Traditional food allergy assessment for anaphylaxis lacks accuracy and accessibility.
• Current anaphylaxis risk assessment methods are expensive and have low predictive value.

Purpose of the Study:

• To develop a machine learning model for patient-specific and allergen-specific anaphylaxis risk assessment.
• To create a quantitative measure of anaphylaxis risk using a peanut allergen score.

Main Methods:

• Utilized large-scale diagnostic data from the Tolerance Induction Program (TIP) immunotherapy.
• Developed a hybrid machine learning algorithm combining two Generalized Linear Models (GLMs) with Bayesian methods for adaptive weighting.
• Analyzed 241 individual allergy assays per patient, organizing data by total IgE subdivision.

Main Results:

• The machine learning model achieved 95.2% recall in predicting peanut anaphylaxis severity in a cohort of 530 children.
• Achieved over 99% Area Under the Curve (AUC) in Receiver Operating Characteristic analysis for peanut allergy prediction.
• The developed algorithm provides a quantitative peanut allergen score for assessing anaphylaxis risk.

Conclusions:

• Machine learning algorithms derived from molecular allergy data offer high accuracy and recall for anaphylaxis risk assessment.
• Further development of algorithms for other food proteins is necessary to enhance clinical food allergy assessment and immunotherapy.