Genome Wide Association Study to predict severe asthma exacerbations in children using random forests classifiers

Area of Science:

• Genetics
• Computational Biology
• Pediatric Medicine

Background:

• Personalized healthcare aims to tailor treatments using genetic and environmental factors.
• Current disease prediction models often rely on limited factors, while complex diseases involve numerous genetic and environmental influences.
• Asthma exacerbations are influenced by multiple small-effect genetic and environmental factors.

Purpose of the Study:

• To test if random forests (RF) can improve predictive models for childhood asthma exacerbations by selecting single nucleotide polymorphisms (SNPs).
• To evaluate the efficacy of machine learning in integrating numerous genetic and environmental predictors for disease prediction.

Main Methods:

• Utilized the Childhood Asthma Management Program (CAMP) cohort, defining severe asthma exacerbations by emergency room visits or hospitalizations.
• Identified top Genome Wide Association Study (GWAS) SNPs using RF importance scores.
• Predicted severe asthma exacerbations using varying numbers of top SNPs (10-320) combined with clinical factors (age, sex, FEV1, treatment group).

Main Results:

• A predictive model using 160-320 SNPs achieved an Area Under the Curve (AUC) of 0.66, outperforming a model with 10 SNPs (AUC=0.57).
• Clinical traits alone resulted in an AUC of 0.54, indicating the importance of both genetic and environmental factors.
• The RF algorithm effectively extracted information from a limited sample size.

Conclusions:

• Random forests (RF) algorithms can effectively identify informative SNPs for predicting asthma exacerbations.
• Machine learning tools, including RF, are valuable for integrating large numbers of predictors in Genome Wide Association Study (GWAS) analyses.
• This approach holds promise for advancing personalized asthma management and prediction.