Multiobjective grammar-based genetic programming applied to the study of asthma and allergy epidemiology

Area of Science:

• Computational biology
• Epidemiology
• Artificial intelligence

Background:

• Asthma and allergy prevalence has significantly increased globally.
• These conditions are complex, influenced by various environmental, psychosocial, socioeconomic, nutritional, and infectious factors.
• Understanding causal mechanisms requires advanced analytical tools.

Purpose of the Study:

• To apply multiobjective grammar-based genetic programming (MGGP) to a dataset of 1047 children.
• To generate predictive models for asthma occurrence.
• To identify models for allergy markers: IgE antibody presence and skin prick test positivity (SPT).

Main Methods:

• Utilized a multiobjective grammar-based genetic programming (MGGP) algorithm.
• Applied MGGP to a comprehensive dataset including environmental, psychosocial, socioeconomic, nutritional, and infectious factors.
• Compared MGGP performance against traditional methods like C4.5, logistic regression, and Random Forest (RF).

Main Results:

• MGGP demonstrated higher average accuracy for asthma prediction compared to C4.5.
• MGGP showed superior accuracy for IgE prediction over logistic regression and C4.5.
• MGGP achieved accuracy comparable to RF while producing more interpretable models.

Conclusions:

• MGGP effectively identified intricate relationships between infections, psychosocial, nutritional, hygiene, and socioeconomic factors in asthma and allergy.
• These complex interactions are often missed by traditional epidemiological techniques.
• The MGGP algorithm is implemented in C++ and publicly available.